WPS4108


       Trade Liberalization, Employment Flows and Wage Inequality
                                                 in Brazil1

               Francisco H.G. Ferreira, Phillippe G. Leite and Matthew Wai-Poi2




Abstract:         Using      nationally      representative,     economy-wide         data,   this   paper
                  investigates the relative importance of trade-mandated effects on industry
                  wage premia; industry and economy-wide skill premia; and employment
                  flows in accounting for changes in the wage distribution in Brazil during
                  the 1988-1995 trade liberalization. Unlike in other Latin American
                  countries, trade liberalization appears to have made a significant
                  contribution towards a reduction in wage inequality. These effects have
                  not occurred through changes in industry-specific (wage or skill) premia.
                  Instead, they appear to have been channeled through substantial
                  employment flows across sectors and formality categories. Changes in the
                  economy-wide skill premium are also important.

Keywords:         Trade liberalization and inequality; employment flows; Brazil.

JEL Codes: D31, F16, J31


World Bank Policy Research Working Paper 4108, January 2007

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the
exchange of ideas about development issues. An objective of the series is to get the findings out quickly,
even if the presentations are less than fully polished. The papers carry the names of the authors and should
be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely
those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors,
or the countries they represent. Policy Research Working Papers are available online at
http://econ.worldbank.org.




1Paper commissioned for presentation at the UNU-WIDER Conference on "The Impact of Globalization
on the Poor in Latin America," Rio de Janeiro, 23-24 September 2006.
2Ferreira and Leite are at the World Bank's Research Department. Wai-Poi is at Columbia University. We
are grateful to Nina Pavcnik and Norbert Schady for kindly sharing their constructed industry concordance
with us; and to Gustavo Gonzaga, Will Martin, Nina Pavcnik, Guido Porto, Martin Ravallion and Erik
Thorbecke for helpful conversations or comments.

                                               2


1.      Introduction


The hypothetical link between "globalization" and inequality (within developed

countries; within developing countries; or between them) has been the subject of a vast

literature over the last 20 years. Even a cursory review of this literature reveals that

"globalization" has been a "catch-all term that is used to describe phenomena as diverse

as trade liberalization, outsourcing, increased immigration flows, removal of capital

controls, cultural globalization and generally faster transmission of international shocks

and trends" (Goldberg and Pavcnik, 2004, p.1). In this paper, we will focus on evidence

pertaining directly to trade liberalization although, in some cases, it is difficult to separate

the impacts of trade liberalization strictly defined from those of outsourcing, or of

increased flows of technical and managerial knowledge.


Most of the literature on trade liberalization in Latin America has focused on Mexico,

Chile and Colombia, and suggests that it has contributed to an increase in inequality (or

at least in the gap between skilled and unskilled wages). Since there was a presumption

that Latin America, like most other developing regions, was abundant in unskilled labor,

this empirical finding appeared to contradict the predictions of the (two countries, two

goods, two factors version of the) Stolper-Samuelson theorem in Hecksher-Ohlin trade

theory. Although the next section briefly reviews some of this literature, this paper

focuses on Brazil, a country where trade liberalization appears to have been inequality-

reducing.


Previously one of the most heavily protected economies in the world, Brazil experienced

an episode of marked trade liberalization between 1988 and 1995. Average nominal

tariffs (weighted by lagged industry imports) fell from 43.4% in 1987 to 13.9% in 1995.

Effective rates of protection fell from 55.8% to 20.0% in the same period. These large

changes in protection had a correspondingly large impact on trade flows:

import/consumption ratios across all manufacturing sectors rose from 15% in 1986 to

31% in 1998. Figure 1 shows the evolution of both tariff rates and trade flows over the

                                                      3


1985-1999 period. It has also been argued that this episode of trade liberalization had a
                         3


substantial impact on labor and total factor productivity growth, with the latter increasing

by 6 percentage points in annual rate terms (Ferreira and Rossi, 2003).


During this period Brazilian inequality, which had been rising until 1989, started a

gradual but persistent decline. Figure 2 shows the long-term evolution of two commonly-

used inequality measures, the Gini and the Theil indices, between 1981-2004, for

household income per capita. The bands around the point estimates denote the 95%

bootstrapped confidence intervals. The Figure highlights the trade-liberalization period of

1988-1995, during which inequality briefly rose (for one year) and then began to fall. The

Theil index fell from 0.75 to 0.71 and the Gini fell by almost two points from 0.61 to 0.59

over this seven-year interval. Both declines are statistically significant at the 1% level. As

shown in Figure 3, inequality also fell in the distribution of hourly wages: the Gini fell by

almost three points, from 0.61 to 0.58 and the Theil fell from 0.78 to 0.72. Both declines

are significant at the 5% level. The economy-wide skill-premium (defined as the ratio of
the wages of skilled workers to those of unskilled workers) fell by 14.3% (see Figure 4).4

Looking only at the skill premium in manufacturing, Gonzaga et al. (2006) find a similar

(15.5%) decline between 1988 and 1995.


Were these two phenomena linked? Did trade liberalization (and other aspects of

increased openness which took place alongside it) cause at least part of the

contemporaneous decline in Brazilian inequality? The literature is somewhat

inconclusive. Focusing on the specific channels of industry wage premia (and industry-

specific skill premia) Pavcnik et al. (2004) find no evidence of any effect from trade

liberalization on the Brazilian wage distribution. More recently, Gonzaga et al. (2006)

argue persuasively that, through the more general channel of changes in the economy-

wide skill premium (as opposed to industry-specific premia), trade liberalization did

3The data reported in Table 1 weigh tariff rates by lagged industry imports. An alternative weighting
scheme (by industry value-added) generates an even more pronounced decline: citing data from Kume et al.
(2000), Abreu (2004) reports a decline in nominal tariffs from 54.9% in 1987 to 10.8% in 1995. Effective
rates of protection fell from 67.8% to 10.4% in the same period.
4We use an education-based definition of skill: skilled workers have 11 or more years of schooling; and
unskilled workers are those with ten or fewer. We return to a discussion of this definition and alternatives
below.

                                              4


reduce wage disparities in Brazil. Although these two studies cover the same period, they

use different data sets and methodologies, which lead them to focus on different aspects

of the same phenomenon.


Despite their differences, both studies focus on workers in manufacturing only. The

manufacturing sector accounted for 16% (13%) of total employment in 1988 (1995) and

there can be no a priori presumption that changes in the skill premium in that sector drive

national wage inequality. During this same period, there has also been a convergence

between urban and rural incomes in Brazil, which is often attributed to agricultural
growth.5 Although agriculture is eminently tradable, it has not to our knowledge been

included in any analysis of liberalization and distribution in Brazil.


This paper seeks to revisit the evidence on Brazil's trade liberalization in a more

comprehensive way. We innovate in four ways. First, we combine the approach used by

Pavcnik et al. (2004) to study trade-mandated changes in industry-specific wage and skill

premia, with a consideration of the economy-wide skill premium on which Gonzaga et al.

(2006) focus, and ask what was the combined effect of the two channels on the wage

distribution in Brazil. Second, unlike previous studies, our analysis covers workers in all

sectors of the economy, including agriculture and services. Third, we explicitly consider

employment responses to the tariff and exchange rate changes that took place over the

period, investigating a channel of impact which is generally acknowledged as potentially

important in theory, but seldom studied in practice. Fourth, we use our estimated models

of wages and employment levels to simulate counterfactual wage distributions which

allow us to decompose the changes in distribution actually observed over the seven years

of trade liberalization into various components ­ some directly attributable to trade

reforms, some which may or may not reflect trade changes, and some which are most

likely independent of trade factors.


Our main findings are, first, that trade liberalization in Brazil did in fact contribute to the

observed reduction in wage inequality in the entire Brazilian economy ­ and not just in


5See Ferreira, Leite and Litchfield (2006).

                                               5


manufacturing. As argued by Gonzaga et al. (2006) - and unlike in Chile, Mexico and

Colombia ­ Brazil's pre-liberalization tariffs (adjusted by import penetration) were

highest for skill-intensive goods. These tariffs fell by more than those for other goods,

leading to a decline in their relative prices. Consistent with the simple Stolper-Samuelson

theorem, this decline led to a decline in skilled wages, relative to those of unskilled

workers, and to a movement of workers away from previously protected industries. As

Pavcnik et al. (2004) found, other channels of impact through industry-specific wage and

skill premia were unimportant.


Second, the decomposition results suggest that: (i) changes in industry-specific wage

premia and skill premia were unimportant. Although changes in tariffs have the expected

sign, the overall effect on the wage distribution was negligible. (ii) The bulk of the trade

impact on the wage distribution occurs through the employment and occupational

reallocation that took place in response to changes in tariffs and relative prices. This

effect accounts for a substantial fraction of the observed reduction in inequality between

1988 and 1995. (iii) Changes in the economy-wide returns to skill ­ which may be at

least partly driven by trade reforms ­ contributed to a further reduction in inequality (as

did changes in other returns). (iv) Changes in the joint distribution of (observed and

unobserved) worker characteristics were inequality-increasing, and partly offset some of

the trade-driven changes in inequality.


The paper is organized as follows. The next section provides a very brief overview of the

literature, focusing on the conceptual channels through which trade reforms affect the

distribution of incomes, and on five or six specific theoretical mechanisms through which

openness has been hypothesized to affect wages and employment. Section 3 describes our

data sets and the methodological approach. Section 4 presents the estimation results for a

set of wage and employment regressions. Section 5 discusses a decomposition of the

changes in Brazil's wage distribution between 1988 and 1995, drawing on counterfactual

distributions constructed on the basis of the models estimated in Section 4. It also

discusses the implications of the wage decomposition for poverty and inequality more

                                                  6


broadly, measured in the distribution of household per capita incomes. Section 6

concludes.


2.       A Brief Literature Review


The literature on the relationship between "globalization" and distribution is now so

extensive that Goldberg and Pavcnik (2004) open their recent survey of the subject by

noting that "the number of literature reviews alone is so large by now, that it seems that a

review of literature reviews would be appropriate" (p.1). Given the existence of two

excellent recent surveys ­ Winters, McCulloch and McKay (2004) and Goldberg and

Pavcnik (2004) ­ we make no attempt at an exhaustive review here. Instead, this section

briefly reviews two sub-themes which are of particular importance for our analysis in this

paper: the channels through which trade reforms affect the distribution of income (and

hence inequality and poverty); and the recent evidence on the distributional effects of
trade reform in Latin America in general, and in Brazil in particular.6


2.1.     Channels of Impact from Trade Liberalization to Household Incomes


If openness to the international economy brings persistent gains in terms of access to new

knowledge and technology; or sustained gains in the growth rate of total factor

productivity, then it is possible that trade liberalization leads to faster long-run economic

growth. Whether or not this is in fact the case has been the subject of a debate, with Sachs

and Warner (1995) and Edwards (1998) among the proponents, and Rodriguez and

Rodrik (2001) leading the skeptics. The current balance of opinion seems to be that

"despite the econometric and conceptual difficulties of establishing beyond doubt that

openness enhances income levels, the weight of experience and evidence seems strongly

in that direction" (Winters et al., 2004, p.78).




6See also Nissanke and Thorbecke (2006) for a thoughtful discussion of the links between globalization
and distribution.

                                                      7


If this is indeed the case, the effect of trade on growth, whether it is mediated through a

faster rate of technology adoption or through greater dynamic efficiency gains from

competitive pressures, is likely to be of first order importance in any understanding of the

relationship between "globalization" and poverty, and policy makers should bear it very

much in mind. Nevertheless, this paper belongs to the (large) strand of literature that

seeks to understand the static or short-term impacts of trade liberalization on the

distribution of incomes. When tariffs (or non-tariff barriers) are reduced or eliminated,

the domestic prices of the relevant goods change. These price changes can affect

household incomes (or consumption) through five main direct channels, namely:

    · Output and input prices. If household members are self-employed, producing,

        trading and consuming different goods, then the first-order approximation to the

        change in their welfare as a result of changes in the price vector is simply
         W = (qi -ci)pi.7 The basic insight is that net producers of those goods
                 
                   i

        whose prices fall as a result of trade reforms lose out, while net consumers gain.

    · Wages. For household members who are employed, the first effect of price

        changes is through the knock-on effect on factor prices, and crucially on the

        individual's wage. The exact transmission mechanisms depend on the degree of

        competition in both factor and product markets, but the benchmark result under

        competitive markets is that as protection declines and relative goods prices move

        against the previously protected good, relative factor prices also move against the

        factor in which the protected good is intensive. This is the well-known Stolper-
        Samuelson theorem in Hecksher-Öhlin trade theory.8

    · Employment. In response to changes in profitability that arise because of the

        aforementioned changes in product and factor prices, the composition of

        production typically changes after trade reforms. Industries whose product prices

        have fallen contract, and those whose relative product prices have risen expand. In




7See Deaton (1997), and discussion in Winters et al. (2004).
8The theorem generates less clear-cut results in a world with more than two countries, goods or factors of
production but, as we shall see, the basic insight from the 2x2x2 version of the theorem remains remarkably
powerful.

                                                      8


        response, there is a reallocation of employment across sectors and, in imperfect

        labor markets, changes in unemployment and/or the size of the informal sector.

    · Consumption Prices. Self-employed workers are not the only people who

        consume. Employed workers, who are affected on the income side through

        changes in their wages and employment sector, are also affected by the changes in

        the relative prices of the goods they consume. Trade models often pay little heed

        to this channel because, if preferences are identical across individuals and

        homothetic, then changes in relative prices will affect all households equi-

        proportionately. But if preferences are not, in fact, homothetic, or if they differ

        across households, then the shares of different goods in their consumption

        bundles will vary, and relative price changes will affect different households

        differently. Under this channel, we also include changes in the quality of

        consumption goods available to consumers, either directly because of differences

        in quality between imported and domestically-produced goods, or because of

        improvements in domestic production as a result of import competition, or of the

        availability of imported inputs.

    · Taxes and Public Expenditures. As tariffs change so, in general, will tariff

        revenues. Although there is much evidence that it is possible to reduce protection

        in a revenue-neutral ­ or even revenue-enhancing ­ manner, if revenues do fall,

        then there will either be a decline in the level of some public service or transfer,
        or a rise in some other tax.9 The incidence of these changes is entirely dependent

        on which expenditures or taxes are altered, and on their (marginal) incidence.


Ultimately, all trade reforms must affect household welfare ­ and its distribution ­

through one of these five primary channels. Much of the discussion in the literature has

focused, however, on the nature of the mechanism through which tariff changes affect

wages and employment levels. A trade reform that lowers tariffs for a number of goods

may affect relative wages and the composition of employment through the standard

9The evidence suggests, however, that it has in many cases been possible to liberalize trade without loss in
revenues. This may occur because non-tariff barriers generate rents for private agents, rather than
government revenue; because some tariffs may initially be above their revenue-maximizing level; or
because some trade reforms occur concomitantly with enhancements in the efficiency of customs agencies.
See Ebrill et al. (1999) and the discussion in Winters et al. (2004).

                                            9


Stolper-Samuelson channel, as described above, but it may also work in a number of

different ways. These have been reviewed in some detail by Goldberg and Pavcnik

(2004), and we provide only a sketch below:


    · There may be changes in the wage premia to specific occupations (as opposed to

       industries), such as managers and highly skilled professionals. (See Cragg and

       Epelbaum, 1996).

    · Trade liberalization may be accompanied by adjustments not only in the national

       composition of production, but in the international composition, with some

       activities being outsourced from developed to developing country locations. One

       hypothesis is that some of these activities are intensive in workers that are

       unskilled by rich-country criteria, but skilled in developing countries. See, for

       instance, Feenstra and Hanson (1996).

    · Technical change that raises firms' relative demand for skill is known as skill-

       biased technical change (SBTC). Whereas in developed countries SBTC is

       typically seen as a competing explanation (vis-ŕ-vis trade openness) for increases

       in skill premia, it has recently been argued that in developing countries, SBTC

       may be spurred on by trade liberalization. See Acemoglu (2003) and Theonig and

       Verdier (2003) for different models of how trade liberalization might lead to skill-

       biased technical change in developing countries. If indeed trade liberalization

       leads to changes in the relative demand for skilled and unskilled workers because

       of induced changes in technology, then this is a separate effect, additional to

       Stolper-Samuelson.

    · Related to the previous two channels is the possibility that greater openness

       changes the quality composition of domestic output. Most goods (shoes, textiles,

       cars or computers) can be produced with very different quality, and there is some

       evidence that greater participation in world trade shifts production towards

       higher-quality goods in (at least some) domestic firms. This may be in response to

       greater import competition, or because exchange rate changes shifts resources

       from non-exporters to exporters. If higher-quality varieties are more intensive in

       skilled workers, this effect too could raise the relative demand for skills in the

                                                     10


         labor force. See Verhoogen (2006) and Shigeoka, Verhoogen and Wai-Poi (2006)

         for evidence from Mexico.

     · Finally, wage levels for observably identical workers are not the same across

         different industries, either because of imperfect competition that gives rise to

         industry rents; or because of compensating differentials; or industry-specific

         skills. It is possible that tariff and mandated price changes affect these

         differentials, in addition to any impact they may have on the economy-wide skill

         premium. See e.g. Goldberg and Pavcnik (2005) on Colombia.


2.2.     The Distributive Impact of Trade Liberalization in Latin America


Contrary to what was found in earlier LDC liberalization episodes, notably in East Asia,

Latin American trade liberalizations during the 1980s and 1990s have been

predominantly contemporaneous with increases in the economy-wide wage skill

premium, which is typically defined as the ratio of wages of skilled workers to the wages
of unskilled workers.10 Although this ratio is not a particularly good measure of wage

inequality, and certainly a very poor indicator of inequality in household incomes, it has
been the focus of most empirical work.11


Evidence of a rising skills gap has been comprehensively established for Mexico, by

Feenstra and Hanson (1995), Cragg and Epelbaum (1996), Feliciano (2001), and Hanson

and Harrison (1999), among others. It has also been documented for Chile by Beyer,

Rojas and Vergara (1999); for Colombia by Attanasio, Goldberg and Pavcnik (2004); and

for Costa Rica by Robbins and Grindling (1999). Given the presumption that developing

countries are abundant in unskilled labor, the first reaction to these results was that it

seemed to contradict the Stolper-Samuelson Theorem, and considerable effort has been

expended in proposing alternative explanations, many of which were described in general

terms in the preceding subsection.

10It matters whether skill levels are measured in practice by education levels (as typically done in studies
based on household survey data) or by production vs. non-production workers (as commonly done in
studies using firm-level data.
11Later in this paper, we investigate how changes in skill premium map into changes in more general
measures of inequality.

                                                 11




Adrian Wood (1997) argued that the simplest version of Stolper-Samuelson may not

apply, because Latin American countries are perhaps abundant in land and natural

resources, rather than unskilled labor; or because of the entry into the international

trading system of countries even more unskilled-labor abundant, such as China and India.

As discussed, Feenstra and Hanson (1995) suggested that part of the increase in the

demand for skill in Mexico was due to outsourcing. Cragg and Epelbaum (1996) argued

that it was driven by the increases in the returns to specific occupations, such as managers

and administrators, who were highly skilled (rather than by returns to all skilled workers

in the economy). Others have argued that greater openness has spurred a process of

technological change that is skill-biased, as also discussed above. Attanasio, Goldberg

and Pavcnik (2004) interpret their finding that increases in demand for skilled workers

were largest for sectors with the largest tariff cuts as supporting the thesis of an

endogenous skill-biased change in technology, that occurs in response to competitive

pressures and to the availability of inputs brought about by greater openness.


Each of these alternative stories ­ occupational rewards; skill-biased technical change;

quality upgrading; outsourcing ­ is plausible, and each is supported by at least some of

the available evidence. But it is also true that an examination of the patterns of protection

in Chile, Colombia and Mexico prior to liberalization reveals that tariffs were generally

higher for industries intensive in unskilled labor (than for skill-intensive industries). In

this case, a fall in the relative prices of these goods, and thus in the price of the factor

they are intensive in, is perfectly in line with the Stolper-Samuelson theorem in the first
place.12


Brazil was an exception, in that effective protection prior to liberalization was higher for

skill-intensive industries. The correlation between tariffs and industry skill-shares in 1988

was mildly positive, and much stronger once tariffs are adjusted by the industry import

penetration rates, which account for differences in the pass-through between tariffs and




12See Goldberg and Pavcnik (2004) for discussion.

                                                      12


prices in different sectors.13 Gonzaga et al. (2006) show that, as tariffs were (partly)

harmonized between 1988 and 1995, the tariff and effective rates of protection (ERPs)

declines for skill-intensive industries were greater than for industries intensive in

unskilled workers. In accordance with Stolper-Samuelson predictions, the relative prices

of skill-intensive goods then fell, as did the relative wages of more skilled workers. Using

mandated wage equations, these authors estimate that the decline in the manufacturing

skill premium over this period was of the order of 25% - larger than the actually observed

15% decline.


In addition, the pattern of labor reallocation was more consistent with a Stolper-

Samuelson effect of trade liberalization, than with a Rybczynski-style effect of increases

in the endowment of skilled labor: the manufacturing employment share of skilled

workers rose by 2.67%, which decomposed into a 3.34% within-industries effect, and a

negative 0.67% percent between industry effect. This contraction in the employment

share of skill-intensive industries would not be expected if the dynamics were driven

primarily by expansion in the endowment of skilled workers in Brazil, but is consistent

with the expected employment reallocation in response to a trade shock.


The evidence presented by Gonzaga et al. (2006) strongly suggests that the Brazilian

trade liberalization of 1988-1995 played some role in the decline of inequality in Brazil

which began during that period. It appears to have done so through the classic channel of

changes in the economy-wide skill premium, in line with the prediction of Hecksher-

Ohlin trade theory, leading to a sizable decline in the wage gap between skilled and
unskilled workers in manufacturing.14


But how important was this change in the skill premium for the actual size distribution of

hourly wages in Brazil? There are two reasons why its importance is far from guaranteed:

one is that skilled and unskilled workers are large and heterogeneous groupings. There is

13See Gonzaga et al. (2006) on the importance of this correction.
14The literature suggests that alternative channels of impact were less important. Pavcnik et al. (2004), for
instance, find no evidence that changes in industry-specific wage premia (or in skill-premia specific to
certain industries) changed in response to trade liberalization.

                                                 13


considerable wage variation within each group, and the two group distributions do

overlap. The second reason is that Gonzaga et al. (2006) consider only manufacturing

workers, which accounted for 16% (13%) of total employment in 1988 (1995). Changes

in their relative position vis-ŕ-vis workers in agriculture (which were also affected by

changes in tariffs) and in services (which were indirectly affected by changes in tariffs,

and also by changes in the exchange rate) may have led to overall changes in wage

inequality which are quite different from those mandated by the Stolper-Samuelson

effects within manufacturing.


In the remainder of this paper we examine two basic questions. First, we seek to place the

changes in wage inequality which can be attributed to trade policy changes in the context

of other changes that were concurrently affecting the wage distribution. Second, and

more specifically, we also seek to quantify the contribution of the trade-mandated

employment reallocation effects to changes in the wage distribution. A third, albeit more

tentative, contribution is to investigate the implications of these trade-driven changes in

the wage distribution for poverty and inequality in the distribution of household incomes

per capita.


3.      Data and Methodology.


3.1     The Data Sets.

The data used in this study come from two different sources. The first of these is the

household survey data, which are used to obtain information on wages, hours of work,

occupations, education levels, age, gender, race and location of workers. We use 11

waves of the Pesquisa Nacional por Amostra de Domicílios (PNAD), fielded by the

Instituto Brasileiro de Geografia e Estatística (IBGE), annually between 1987 and
1999.15 The PNAD is a nationally representative household survey, with a stratified and

clustered sampling design which ensures coverage of rural and urban areas in every state

of the federation, except for the rural areas of Acre, Amapá, Amazonas, Pará, Rondônia




15There are eleven waves because the survey was not carried out in 1991 or 1994.

                                                        14


and Roraima.16 Sample sizes varied somewhat from year to year, around approximately

300,000 individuals per year.


For our wage analysis, we considered all workers aged 15-65 who reported positive

earnings during the survey's reference week. Since we are interested in quantifying the

importance of trade and openness-related changes in explaining overall changes in the

wage distribution, we include all workers, in agriculture, industry and services, regardless

of formality or own-account status. Effective sample sizes for this analysis were 107,195

workers in 1998 and 123,455 workers in 1995. This is an important difference between

our analysis and those of Gonzaga et al. (2006), or Pavcnik et al. (2004), who focus
exclusively on workers in manufacturing.17 The wage definition is hourly wages,

calculated as a quarter of the monthly wage, divided by the number of hours worked on

the average week. All monetary values are inflated to September 2004 prices, using the

INPC deflator with the Corseuil and Fogel (2002) adjustment to the 1994 index. See

Ferreira, Leite and Litchfield (2006) for the full deflator in each PNAD reference month.


As all PNAD-based studies, we use years of schooling as our measure of a worker's
skill.18 In earnings regressions, we group workers into nine schooling groups: zero years;

1-3 years; 4 years; 5-7 years; 8 years (completed primary); 9-10 years (some high

school); 11 years (completed high school); 12-14 years (some university); 15+ years

(completed university). We also use this variable to construct a dichotomous skill

indicator, classifying workers with 0-10 years of schooling as unskilled, and those with

11 or more (completed high school and above) as skilled. Earnings regressions were also

estimated with an alternative indicator, which classified only workers with 15 years of



16These rural areas broadly correspond to the Amazon rainforest, which was excluded from PNAD
sampling until 2003. Census data suggests that these areas account for 2.3% of the Brazilian population.
17The sample in Pavcnik et al. (2004) is, in addition, based on the Pesquisa Mensal de Emprego (PME) and
is thus only representative of the country's six main metropolitan areas.
18Use of education to define a worker's skill is in fact common to most household or labor-force survey-
based studies. Articles relying on firm-level data typically classify production workers as unskilled, and
non-production workers as skilled. Although Slaughter (2000) shows that the two definitions do not seem
to lead to very different conclusions in the US, Gonzaga et al. (2006) show that the distinction does matter
for Brazil. Like the latter authors, we feel that the education classification is more appropriate in the
Brazilian case, since trade liberalization and outsourcing of support activities led to considerable changes in
the employment levels of non-productions workers during the period.

                                                    15


schooling or more as skilled, and all results were qualitatively robust. Since workers with

completed university accounted for 4.5% of the labor force in 1988; while workers with

complete high school and above represented 16.4%, we chose the latter classification as

more meaningful for Brazil. See Gonzaga et al. (2006) for a similar discussion of this

classification.


The second data set used for this study comprises the trade-related variables for 22

industries, across the 1987-1999 period. Six trade-related variables are used: nominal

tariffs and effective rates of protection come from Kume et al. (2000), as reported in

Abreu (2004); import penetration and export shares by industry come from Muendler

(2003); and import-weighted and export-weighted industry-specific exchange rates were

constructed by the authors, based on the methodology suggested by Goldberg (2004), and

using data from the World Bank's World Development Indicators and the UN's
COMTRADE.19 Table 1 presents initial (1988) and terminal (1995) values for each of
these variables for the 22 industries into which we have grouped Brazilian firms.20 As in

Pavcnik et al. (2004), our use of household survey and trade data with different industry

definitions necessitated a concordance between the various datasets, mapping the more

disaggregated industry classifications in the trade data to the broader PNAD
classifications. In addition to the standard Nível 80 to Nível 100 concordance,21 we

developed concordances of our own. The main one is based on (but not identical to)
Pavcnik et al. (2004), and is presented in Table 2.22 A more detailed description of the

steps taken to clean, construct and match the data is included in the data appendix.




19As Goldberg (2004) notes, "At the national level, analyses of exchange rate moves often rely on
aggregate trade-weighted exchange rates. However, aggregate indexes can be less effective than industry-
specific indexes in capturing changes in industry competitive conditions induced by moves in specific
bilateral exchange rates." (p.1)
20Although we use data until 1999 in our estimation stage, the key initial and terminal years for our wage
decomposition analysis are 1988 (at the onset of trade liberalization) and 1995 (the year in which it was
completed.
21Available on Marc Muendler's excellent website of Brazilian data resources
(http://www.econ.ucsd.edu/muendler/html/brazil.html). Nível 80 and 100 are official Brazilian industry
classifications.
22We are grateful to Nina Pavcnik and Norbert Schady for graciously making the details of their
concordance available to us. A further concordance between Nível 100 and the trade categories used by
Kume et al (2000) was also constructed, but is too detailed to present here.

                                                    16


3.2.    Methodology.

Since the objective is to understand and quantify the role of trade-induced changes in the

wage distribution, in relation to the overall observed changes, we combine an extended

version of the two-stage estimation framework which has recently been used to

investigate the effect of trade reforms on wage premia in a number of settings, with a

more general decomposition of changes in the entire wage distribution. Following

Pavcnik et al. (2004), our first stage regresses log hourly wages (wij) on a vector of

worker i's characteristics (including sex, race, experience, education, residential region,

urban/rural status, household headship status, and formality status); a vector of industry j
indicators (Iij); and a set of interactions between industry indicators and skill category:23


ln wij = Xij + Iij * wpj + (Iij * Sij )spj + ij                                                    (1)



Equation 1 is estimated separately for each year in the data set, from 1987 to 1999. In

addition to the wage equation, our first stage also includes a model of employment for

each year in the sample, where an individual's occupation is regressed on a similar set

(Zij) of personal characteristics, as well as whether or not he or she has children, and the

spouse's occupational status. Given the polychotomous nature of the occupational choice,

this relationship is estimated with a multinomial logit model:



Pr{j = s}= Ps (Zi,) =            eZi s
                                                                                                   (2)
                           eZi s + eZi j
                                   
                                   js




In equation (2), there are ten possible occupational choices (j), corresponding to

inactivity; unemployment; self-employment; employer status; and formal or informal

employment in each of three broad sectors: agriculture, manufacturing and services. The

full specification of models (1) and (2) is presented in the next section, alongside with

results.


23Recall, skilled workers are defined here as having completed high school or better (11+ years of
education).

                                               17




In the second stage, three sets of estimated coefficients from the first stage are pooled

over time and regressed on a set of trade-related industry characteristics. The dependent

variables in this stage are: (i) the industry premia coefficients from (1), wpjt; (ii) the

industry-specific skill premia coefficients from (1), spjt; and (iii) the constant terms 0jt

for each occupation from (2). Each of these variables vjt ={wpjt; spjt; 0jt) is regressed (in

first differences) on a set of trade-related variables (Tjt), including industry- specific

effective rates of protection, import- and export-weighted exchange rates; import

penetration and export shares:


vjt = Tjt +jt                                                                                  (3)



In addition to estimating a two-stage employment model, which we have not seen in the

trade literature, there are other differences between these estimations and those reported

in Pavcnik et al. (2004). Our data comes from a nationally representative household

survey (the PNAD), and include workers in agriculture and non-tradable industries (for
which industry-specific exchange rates can be constructed,24 and affect relative prices).

Rather than relying on the manufacturing sector in the six largest metropolitan areas of

Brazil, our sample is therefore much more broadly representative of the country. Possibly

as a result, some of our estimation results are different than those in Pavcnik et al. (2004),

and we discuss them briefly in the next section.


But the main purpose of estimating equations (1)-(3) is to use them in the decomposition

of all changes in the full wage distribution between 1988 and 1995. Following Juhn,

Murphy and Pierce (1993) ­ henceforth cited as JMP ­ one can decompose the difference

between the wage distribution prior to the liberalization (say, in 1988) and the wage

distribution afterwards (say, in 1995) into three components: one due to changes in

observed worker characteristics (X); one due to changes in the returns to those

characteristics (measured by regression coefficients ) and a final one due to changes in



24We use economy-wide import- and export-weighted exchange rates for the non-tradables sector.

                                                       18


the distribution of residuals (). Writing the distribution function of the residuals of

equation (1) as it = Ft (it ), a standard JMP decomposition would proceed as follows.25


After estimating earnings regressions (like (1)) in both initial and terminal years:

ln wi 88 = Xi 88 + F88 (i ) and
                           -1
               88               88


ln wi 95 = Xi 95 + F95 (i ),
                           -1
               95               95

one would simulate two counterfactual wage distributions, as follows:
ln wi = Xi 95 + F88 (i )
     a                   -1                                                                             (4)
             88                88

ln wi = Xi 95 + F95 (i )
     b                   -1                                                                             (5)
             88                88




The difference between wage distributions G w88 and G       ( )           (w )
                                                                             a   would be interpreted as

being due to differences in returns between the 1988 and 1995. The difference between

G wa
   ( )    and G wb ( )     would be due to changes in the distribution of (or returns to)

unobservable worker characteristics. Finally, the difference between G w95 and G wb       ( )          ( )

would be due to changes in the joint distribution of observed worker characteristics (and
their joint correlation with the residuals).26


Using our estimates of equations (1) ­ (3), we construct an expanded set of such

counterfactual wage distributions, which also form an exact decomposition of the
observed change between 1988 and 1995.27 We construct six counterfactual wage

distributions, chosen to shed light on different channels of effect from the trade-

liberalization. Not all of the channels discussed in Section 2 are covered, but the key

wage and employment channels are addressed.


25We say "a standard JMP decomposition" because there are variations on the basic theme. Alternative
orderings could be used, or average coefficients instead of final year coefficients, etc. The specific
interpretation varies, but the essence of the insight is the same.
26As noted by JMP, and in the closely related work of DiNardo et al. (1996) and Bourguignon Ferreira and
Lustig (2004), this is an accounting decomposition. Changes in the return structure are clearly causally
related to changes in the distribution of characteristics (in more than one way). The decomposition exercise
does not disentangle these causal relations, but provides a description of the observed changes.
27As in the original JMP decomposition, and indeed in any generalized Oaxaca-Blinder decomposition, the
order of the simulations matter: the decomposition is path dependent. See Lemieux (2002) and
Bourguignon et al. (2004) for discussions.

                                                    19




Departing from equation (1), our first counterfactual is:
ln wij = Xij  88 + Iij * wpsj + (Iij * Sij )sp88 + F88 (i )
    1        88         88            88    88            -1                                           (6)
                                                   j           88

where wpsj = Tj -Tj  wp and ^wp are the estimated coefficients in the second-stage
                  ( 95     88)^

regression (3) for the industry wage premia. This first simulation therefore corresponds to

changes in the wage distribution due only to those changes in industry wage premia

which are mandated by changes in the exogenous trade variables included in the second-

stage equation (3). The second counterfactual is:


ln wij = Xij  88 + Iij * wpsj + (Iij * Sij )spsj + F88 (i )
     2       88         88            88    88           -1                                            (7)
                                                               88

where spsj =     (T 95    88)
                   j  -Tj ^sp      and ^sp are the estimated coefficients in the second-stage

regression (3) for industry-specific skill premia. The third counterfactual is


ln wij = Xij  88 + Iij * wpsj + (Iij * Sij )spsj + F88 (i )
     3       88         s            s    88           -1                                              (8)
                                                             88


where Iij is a counterfactual vector of occupations, constructed by substituting
            s



0j = Tj -Tj ^ into the occupational multinomial logit in equation (2), and then
  s    (   95     88)
                       0

using it to predict the corresponding counterfactual distribution of occupations.28 This

third simulation therefore corresponds to the overall effect of industry premia, industry-

specific skill premia and employment changes mandated by the second-stage trade

variables, under the maintained functional form assumptions in (3).


The power of the preceding counterfactuals to simulate the changes in the wage

distribution that arise from trade reforms depends entirely on the ability of the linear

second stage equations to identify the impact of changes in tariffs and exchange rates on

wage premia and employment probabilities. They also miss, so far, a key theoretical

channel through which a trade liberalization is likely to impact on wage differences in the

28As noted above, the ten occupational categories specified in equation 2 were more aggregated than the 22
industries for which wage and skill premia are estimated. In constructing G(w3), workers whose predicted
occupations differed from those observed in 1988 due to changes in 0j were allocated to specific industries
(within the broad sector to which they were mapped by the multinomial logit) by random draws with
probabilities derived from the 1995 employment distribution across industries.

                                              20


economy, namely changes the economy-wide skill premium. After all, the wpj and spj

coefficients capture only changes in industry specific remuneration rates, controlling for

changes in the average returns to education. Our fourth counterfactual is therefore:


ln wij = Xij  + Iij * wp95 + (Iij * Sij )sp95 + F88 (i )
    4      88  s   s            s    88          -1                                     (9)
                          j                j          88

where  = ed ;-ed . The difference between (9) and (8) is twofold: the industry and
         s   {  95  88 }
industry-specific skill premium coefficients mandated by the second stage are replaced

with those estimated in 1995; and the economy-wide returns on schooling coefficient is

replaced with its 1995 value. This simulation therefore corresponds to a "more generous"

estimate of the "price effects" of trade liberalization, in which the full changes in returns

to education and to industry membership ­ rather than only those mandated by the second

stage ­ are included. Although the main channel through which trade reforms might

affect ed is the Stolper-Samuelson effect of reduced protection in skill-intensive
          95


industries, there may be other channels too. If one is prepared to accept that skill-biased

technical change in Brazil, or skill-demanding changes in the quality composition of

domestic output, are endogenous to trade liberalization, as discussed in Section 2, then

one may come closer to the view that all changes in the returns to schooling between

1988 and 1995 are, in some way or another, related to trade. Be that as it may, we see this

particular counterfactual wage distribution as a generous estimate of the joint wage and

employment effects of trade on the wage distribution. The true contribution of trade

liberalization to changes in Brazilian inequality is likely to lie somewhere between the

changes accumulated up to equation (8) and those corresponding to (9).


Three remaining steps (and two counterfactual distributions) complete the decomposition

­ and represent changes that are less likely to have been driven by trade reforms. The first

of these (in equation 10) computes the additional changes in the structure of returns to

observed characteristics (like experience, location, race, gender, etc.). The second (in

equation 11), brings in the distribution of residuals from 1995, in a rank-preserving

transformation:

                                                  21


ln wij = Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F88 (i )
      5      88        s              s   88           -1                                        (10)
                              j                  j           88

ln wij = Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F95 (i )
      6      88        s              s   88           -1                                        (11)
                              j                  j           88




Finally, the difference between (11) and the equation estimated for 1995:
ln wij = Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F95 (i )
      95      95        95             95    95           -1                                     (12)
                                j                  j            95

corresponds to differences in the joint distribution of observed characteristics between

1995 and 1988 (except for the changes in occupational structure due to trade, which had

been predicted in equation 8). This step also accounts for changes in the correlation

between the observed characteristics and the residual terms, including any changes in

selection into the labor force.


Only equation (1), for 1988, and equation (12), for 1995 are estimated on observed data.

Equations (6)-(11) give rise to simulated wage distributions G             (w )
                                                                              1  to G w6 . In Section
                                                                                       ( )

5, simulation results are presented in two ways. First, a number of inequality indicators

are computed for each counterfactual distribution, so that we can decompose the

observed changes between 1988 and 1995 into the components corresponding to each

counterfactual. Second, we can see a fully disaggregated picture by plotting the observed
wage growth incidence curve between 1988 and 1995, g(p) = w95 - w88 w88 , and       p      p     p

presenting        each      intermediate       counterfactual        growth       incidence       curve:
g s (p) = wsp - w88 w88 , s = 1, ...,6.29
                   p     p




4.       Estimation Results


Before we turn to the simulation results in the next section, this section presents the

estimation results on which they build. Table 3a presents the main first-stage results for

the wage equation (1), while Tables 3b and 3c report the industry-specific wage and skill




29The "wage growth incidence curve" is an application to the distribution of wages of Ravallion and
Chen's (2003) concept of growth incidence curve, which those authors originally defined on the
distribution of household incomes or expenditures.

                                            22


premia coefficient estimates respectively. Each model was estimated for each year in the

1987-1999 interval for which data were available. The results in Table 3a are in line with

existing analysis of the Brazilian labor market (see e.g. Ferreira and Barros, 1999). There

are large and significant returns to education, and smaller and concave returns to

experience. Measured with respect to zero years of schooling, returns to education fell

consistently over the period. This decline was most pronounced for intermediate

education categories (4-10 years of schooling). Returns to experience have also fallen.

There is a substantial male wage premium, which has also been declining. In contrast,

racial premia of both whites and Asians with respect to Afro-Brazilians have persisted or

increased. Controlling for other observed characteristics, employers, the self-employed

and formal employees all earn more than informally employed workers. Metropolitan and

urban location premia vis-ŕ-vis rural workers of identical characteristics have also fallen

over the period, though they remain statistically significant. All specifications also

include industry and interacted industry-skill indicators, the coefficients on which are

reported in Tables 3b and 3c. These coefficients are then pooled and regressed on a

vector of trade variables in the second stage.


Table 4a reports the first-stage results for the employment model in (2), as marginal

effects of unit changes in each independent variable, with all other variables held at their

mean values. These estimates are also mostly in line with expectations. Workers with

more experience and education are less likely to be employed in agriculture or

informally. In addition, those with higher education are also more likely to be employers,

as are whites and Asians, and to work in the formal non-tradables sector, presumably

often as professionals. We also see some evidence over time of more females entering

the labor force. While men are more likely to be working, the male coefficient falls

between 1988 and 1995 across most industries. The industry constants, which are pooled

and used in the second stage, are summarized in Table 4b.


One concern that is typically voiced with respect to multinomial logit models such as (2)

is that they assume that the odds ratios of any two possibilities (pj/pk) are independent of

the number and nature of alternative outcomes. This is known as the independence of

                                                   23


irrelevant alternatives (IIA) hypothesis. When alternative polychotomous discrete choice

models that do not make this assumption, such as multinomial probits, are unstable or

display convergence problems, one can test for the validity of the assumption using, for

instance, the Hausman specification test which, in essence, tests for the stability of

parameter estimates as alternative outcomes are excluded from the model. This test failed

to reject the null hypothesis (that IIA is satisfied) for 8 out of our 9 outcome categories

for 1995. Although results were poorer for 1988, with four rejections of the null, the

overall picture is not one of overwhelming rejection of the Multinomial Logit
specification.30


Table 5 reports the second-stage regression of industry wage premia on effective rates of

protection, import penetration, export shares, and import- and export-weighted real

exchange rates. All specifications are in first-differences. Import penetration rates and

export shares are entered only in lags, so as to reduce possible simultaneity concerns. The

basic argument for treating changes in effective rates of protection ­ the main variable of

interest ­ as exogenous is the same as in Attanasio, Goldberg and Pavcnik (2004) and

Pavcnik et al. (2004): trade reforms in Latin America in the late 1980s and early 1990s

arose as a response to becoming GATT / WTO members, or to a central policy decision

to comply with previously negotiated rules. "This reflects the government's objective to

reduce tariffs across industries to more uniform rates negotiated with the WTO.

Policymakers accordingly cater less to special lobby interests, so that tariff declines in

each industry are proportional to the industry's pre-reform tariff levels (...) alleviating

concerns about endogeneity at least in the economic sense" (Goldberg and Pavcnik, 2004,

p.4.)


Eight different specifications are presented in Table 5. While ERPs are insignificant

when differences are entered contemporaneously, they become robustly statistically

significant in first lags (specifications 4 ­ 8). The coefficients on lagged ERPs have the

expected positive sign and are statistically significant at the usual levels, suggesting that

larger declines in protection were associated with larger declines in the industry wage


30Details from the Hausman specification test are available from the authors on request.

                                                        24


premium over this period in Brazil.31 Nevertheless, the estimated size of the impact is

small: using the ERP coefficient from specification 8, the fall in average ERP from

51.4% in 1988 to 20.0% in 1995 (with all other variables held at their mean values)

would result in a 1.6% decrease in average industry wage premium.


Although theory suggests that the pass-through of tariffs to product prices, and thus to

wages, is mediated by the sector's import penetration (see Gonzaga et al., 2006), import-

penetration does not appear to be important in mediating the effect of tariffs on industry-

specific wage premia, as shown by the insignificant coefficients on the interaction terms.

The same is not true of (import-weighted) exchange rate effects, which have the predicted
(negative) sign32 when interacted with lagged import penetration: as the currency

appreciates and imports become more competitive for a particular industry, wage premia

in that industry decline. When the RER is export-weighted and interacted with lagged
export-share, the effect is positive and significant.33


Table 6 presents the second-stage results for the regression of industry-specific skill

premia on the same set of trade-related variables. For industry-specific skill premia, once

the economy-wide returns to skill are controlled for in the first stage, there are no

particular theoretical predictions, and we find that the coefficients on ERPs are

insignificant across all specifications. Interestingly, however, we find a fairly robust

pattern of negative and significant coefficients on lagged import penetration, suggesting

that skill premia were falling for those industries where the growth in import penetration

was largest. We know that these were largely skill-intensive industries that were most

highly protected prior to 1988, as seen in Table 1. This movement in industry-specific

skill premia is therefore consistent with the decline in the economy-wide skill premium


31Although Pavcnik et al. (2004) expected these results ("The models predict a positive association
between industry tariffs and wages, so that declines in industry tariffs lead to proportional declines in
industry wages", p.321), they did not find them. This may have been due to their use of the less
representative PME data, or to specifications that did not include lagged ERPs. Lagging ERPs accounts for
the fact that the effects of reduced protection on industry premia may take time to flow through. Our
protection results are, however, consistent with those for Colombia (Attanasio, Goldberg and Pavcnik,
2004), and Mexico (Revenga, 1997).
32An increase in our exchange rates means an appreciation in the currency.
33This likely reflects the fact that an increase in an industry's export share would be expected to increase
that industry's wage premium.

                                                   25


which was documented in Gonzaga et al. (2006), and which we also observe. Controlling

for the growth in import penetration, there appears to be some evidence that a stronger
currency increases the skill premium.34


Table 7 reports the second-stage results for the regression of industry participation

constants from the employment multinomial logit model on the same set of trade-related

variables. These results are somewhat harder to interpret, since the ten occupational

categories used in the estimation are much more aggregated than the 22 industries used in

the previous two tables, and are basically at the agriculture, manufacturing and services
level.35 Partly as a result, there is a counterintuitive negative sign on lagged ERPs, which

suggests a (conditional) movement towards the industries experiencing the greater

declines in protection. This result is explained by a movement towards the tradable

sectors (particularly agriculture) of the reference category of workers in the employment

model. Once we look at the unconditional pattern of employment changes for all workers

at the disaggregated industry level, in Figure 5, we observe the expected positive

correlation: employment levels seem to have fallen by more in industries experiencing

larger declines in protection.


The only other statistically significant result in Table 7 is easier to interpret: lagged

export shares are positively correlated with conditional increases in employment,

suggesting that industries that succeeded in increasing their exports suffered smaller

declines in employment than others.


Taken together, these results paint a mixed picture. The signs and significance are

broadly ­ if not wholly ­ consistent with theoretical expectations from models that

feature barriers to labor movement across sectors. Larger falls in protection and an

exchange rate that makes imports more competitive domestically imply lower industry


34This may be evidence of quality upgrading in the face of increased import competition.
35Both formal and informal agriculture categories are regressed on trade variables for the agricultural
industry. Formal and informal manufacturing use manufacturing trade variables averaged over the various
manufacturing industries, using lagged imports as weights. Formal and informal nontradables, employers
and the self-employed categories are assigned zero ERPs, import penetration and export share, but use
economy-wide import- and export-weighted real exchange rates.

                                             26


wage premia. More exports in an industry are associated with increased wage premia and

employment. Greater import penetration is associated with falls in wage premia which

are greater for more skilled workers (who work mostly in industries that suffered the

largest increase in penetration). These results are perhaps more in line with theory and

with other accounts of trade liberalization in Brazil than previous attempts at estimating

these relationships in Brazil, notably by Pavcnik et al. (2004).


Yet, they leave much to be desired. Empirically, the R2 of each second stage regression

is never higher than 0.12. For the wage equations, they are never higher than 0.04,

suggesting that, however economically important the joint variation in the trade variables

may be, it accounts for a small share of the observed variation in wage premia across

industries and over time. Conceptually, these wage regressions focus on only one of the

five mechanisms through which trade reforms are thought to influence changes in wages

in developing countries (which were reviewed in Section 2), namely changes in industry

specific wage and skill premia.


However, if workers can move across industries over the medium-run, and if market

imperfections in labor and product markets are not particularly severe, then it is likely

that the main effects of the changes in protection observed in Brazil over this period

manifest themselves through (i) worker reallocation across industries and (ii) changes in

the economy-wide skill premium. This prediction would accord with the Stolper-

Samuelson theorem, but might also be consistent with economy-wide trade-induced

SBTC or quality ladder models. In the next section, we turn to the full decomposition of

wage changes in Brazil between 1988-1995, in an effort to place the changes implied by

the second stage regressions reported in Tables 5-7 into context ­ both vis-ŕ-vis other

changes that may be associated with trade channels, and vis-ŕ-vis other economic

processes that are less likely to be driven by trade reforms.


5.     Decomposition Results


5.1    The Distribution of Hourly Wages

                                              27


Table 8 summarizes the results of the decomposition described in Section 3.2. It presents

four measures of inequality for the 1988 and 1995 hourly wage distributions in Brazil, as

well as for the six intermediate counterfactual distributions previously described. The
measures are the 90th/10th percentile ratio; the mean log deviation (also known as GE(0),

or Theil-L index); the Theil-T index (or GE-1) and the Gini coefficient. Figures 6-11 plot

the observed wage growth incidence curve (WGIC) between 1988 and 1995, as well as

different counterfactual WGICs, each corresponding to one of the counterfactual

distributions listed on Table 8. The figures provide a full-distribution, disaggregated

decomposition of wage changes.


The differences between G(w1) and G(w88), which correspond to the impact of the trade-

mandated     changes     in   industry-wage    premia,     are  economically     insignificant:

counterfactual inequality measures hardly move, and the counterfactual WGIC in Figure

6 remains very close to the x-axis. Despite statistically significant coefficients on the

tariffs in the second-stage estimation described in the previous section, it appears that

changes in the wage distribution due to industry wage premia between 1988 and 1995

were immaterial. The same is true of changes in industry-specific skill-premia, which are
incorporated into G(w2), in Figure 7. Thus, although our second-stage regression

coefficients are statistically significant (while theirs are not), we reach the same essential

conclusion on the economics of these impacts as Pavcnik et al. (2004): trade

liberalization did not affect the Brazilian wage distribution through industry specific

premia.


But when relative prices and wages change, firms and industries contract and expand in

response. Workers flow across sectors and industries, and their movement is highly
selective (on observed and unobserved characteristics). The difference between G(w2)
and G(w3) is meant to capture those occupational (employment) changes which took

place in response to changes in trade-related variables (as predicted by the second-stage

regressions). These counterfactual changes are much larger than those associated with
industry-specific wage and skill-premia. All four inequality measures for G(w3) move

closer to their 1995 values: the difference in inequality between this simulation and 1988

                                                     28


ranges between 51% of the 1995-1988 difference (for p90/p10) and 76% (for the Theil ­

T). Figure 8 reveals that the bulk of the underestimate is due to the bottom of the
distribution: whereas G(w3) generates a remarkably good prediction of changes in the
wage distribution from the 20th percentile upwards, it considerably underestimates gains

for the bottom quintile.


Allowing for changes in the economy-wide returns to education (and thus in a flexible

version of the economy-wide skill premium) contributes to a further reduction in

inequality, which now in fact overshoots the 1995 targets (for three of the four
measures).36 Consistent with the decline in returns to higher levels of schooling, this

simulation does not affect the bottom of the distribution much, but lowers counterfactual

incomes in the middle and at the top (Figure 9).


It is harder to attribute these changes to trade reforms, since this counterfactual imports

observed 1995 coefficients (on education, as well as on industry dummies and industry

skill premia), rather than those mandated by the second-stage. The bulk of the difference

between G(w3) and G(w4) is due to ed which, by its very nature as an economy-wide
                                               95


vector of returns, does not vary by industry and can not be estimated in a second stage.

But the fact that it can not be included in a Pavcnik et al.-style second stage does not

mean that it does not reflect trade changes. In fact, as discussed above, if output and labor

markets are reasonably well-functioning, a number of theories of trade would predict an

important effect of trade liberalization on this coefficient. The Stolper-Samuelson effect

would predict a decline in the economy-wide skill premium, and thus an inequality-

reduction from importing ed . Trade-induced SBTC, as well as most versions of the
                                    95


outsourcing or quality ladder stories would imply an increase in the demand for skilled

workers, and thus an increase in inequality from ed .         95




36This result is reminiscent of the Gonzaga et al. (2006) finding that trade-mandated changes in the
(economy-wide) skill premium were larger than those actually observed.

                                            29


On the other hand, changes in ed clearly also reflect other economic and demographic
                                  95


changes, notably changes in the supply of skilled workers. As shown in Figure 4, there

was indeed some growth in the skills of the Brazilian labor force over this period, with

the share of skilled workers rising from 20% to 24%. All in all, this decomposition

method does not allow us to separate these changes in inequality into components due to

each of the alternative trade hypotheses, and a component due to changes in the supply of

skills (or to technology changes unrelated to trade). All that can be said is that the net

effect suggests that the Stolper-Samuelson channel and the effects of increased skill

supply (which go in the same direction) seem to have outweighed any effects of SBTC or

quality compositional changes that might have occurred as a result of trade liberalization.


The two final steps in the decomposition incorporate changes in the remainder of the
earnings regression coefficients (the other elements of ) to generate G(w5); and a rank-
preserving transformation in the distribution of residuals to generate G(w6). Finally, the
differences between G(w6) and G(w95) are residually due to changes in the joint

distribution of observed characteristics (and in their correlation with the unobserved,

including selection). Although effects of trade on the returns to experience, or to

unobserved skills, can not be ruled out, these are not channels on which the literature has

focused. Accordingly, we interpret these remaining changes as those which are not

attributable to trade effects. Changes in non-education returns are mildly equalizing, and

poverty-reducing. Changes in the distribution of residuals contribute to lower incomes in

the middle of the distribution and higher incomes at the very top. The net effect is mildly

inequality increasing (except for p90/p10). The effect of changes in the distribution of

observed characteristics further lowers incomes in the middle of the distribution and
raises them above the 75th percentile.


5.2.    The Distribution of Household Income per Capita
Once the six counterfactual wage distributions G(w1) to G(w6) have been simulated, it is a

simple matter to create the corresponding counterfactual distributions of household

income per capita. Household identifiers link each worker in our data set to a particular

household, and information is available on all of its other sources of income (subject to

                                             30


the usual misreporting and measurement issues in an income survey like the PNAD). It is

therefore possible to simulate the impact of these counterfactual changes in wages on the

distribution of household incomes, and on the inequality and poverty levels associated

with it. These results are reported on Table 9, and in Figures 12-17, for the same

inequality indices used so far and for the three standard FGT poverty measures. We adopt

a relative poverty line of R$87.55 in 2004 prices, which corresponds to 50% of the 1988

median household per capita income.


Before discussing these results, it is important to recognize that their limitations are even

greater than those for the wage distribution decompositions analyzed so far. In addition to

the same caveat about path dependence, now the absence of general equilibrium effects

extends to any indirect impacts of trade (or any other changes) on family composition, or

on the occupational decisions of household members other than spouses. There are also

important changes in other, unrelated policy parameters, such as the real value of pension

payments and other transfers, which are consigned to the residual ­ which is therefore

larger than in the decomposition described in Table 8.


As in the hourly wage distribution, trade-mandated changes in industry wage premia and

industry-specific skill premia have very limited effects (although they are somewhat

more inequality increasing, suggesting that the workers hardest-hit by wage declines in

contracting industries belonged to poorer households: Figures 12 and 13). The biggest

impact, as before, comes in the transition from the second to the third counterfactual

(Figure 14). The changes in occupations across the distribution which occur in response

to changes in the occupation-specific terms in the multinomial logit model are vastly

poverty- and inequality-reducing. They contribute a decline of six points in the headcount

index, actually overshooting the observed decline. Inequality measures move very close

to the observed 1995 values, and the counterfactual growth incidence curve

corresponding to these "full trade effects" lies quite close to the observed GIC (1995-

1988).

                                                   31


Allowing for changes in the returns to education ed in simulation 4 contributes to a
                                                                 95


further reduction in inequality, as in the wage distribution, and to an under-prediction of

all, but particularly the highest, incomes (Figure 15). That is partly corrected by allowing

the other earnings regression coefficients (including the constant) to take their 1995

values (Figure 16). Finally, replacing the earnings regression residuals with a rank

preserving transformation of the 1995 residuals leads to column HPCI (6) in Table 9, and

to the counterfactual GIC in Figure 17.


The underestimates in poverty and inequality implied in this final counterfactual reflect

two main factors. First, there were substantial changes in labor force participation and

informality over this period which were unrelated to trade and are thus not captured by

the simulation. Second, there were also important changes in the incidence of non-labor

incomes, with a decline in the real value of minimum-wage linked pensions at the bottom

of the distribution, and an increase in the real value of retirement earnings at the top. Both

of these trends, which were documented in Ferreira and Barros (1999), help account for

the difference between the counterfactual and the actual GICs in Figure 17 (and between

the two bottom lines of Table 9).


These earnings-based simulations are not the most suitable way for understanding

differences between full household income distributions. The extended version of this

approach which is described in Bourguignon et al. (2004) would be much more
appropriate.37 The point of this subsection, then, was merely to point out that the links

between trade ­ and, in particular, trade-mandated employment flows across sectors and

formality status ­ and wage inequality do appear to carry through to the changes we have

observed in the distribution of HPCY in Brazil over this period, including a substantial

part of the observed poverty reduction.



37Yet another alternative approach would be to decompose (both the actual and counterfactual) changes in
inequality between 1988 and 1995 into "horizontal" and "vertical" components. The former reflect
differences in the income changes affecting households at similar initial levels of welfare, while the latter
captures differences in (conditional mean) changes across different levels in the initial welfare distribution.
See Ravallion and Lokshin (2004) for a specific decomposition, applied to a (counterfactual) trade reform
in Morocco.

                                            32


6.     Conclusions


Using a nationally representative sample of workers in all sectors of the economy, this

paper has sought to quantify the impacts of the 1988-1995 trade liberalization episode on

the Brazilian wage distribution. Our results confirm previous findings that changes in

industry wage premia and industry-specific skill premia did not meaningfully contribute

to changes in the distribution of hourly wages. Trade reforms did contribute to the

observed reduction in inequality, but this happened through other channels. Chief among

them were trade-induced changes in employment levels across sectors, industries and

formality categories (formal, informal, self-employed, employer). The reallocation of

workers that our model predicts to have arisen from changes in levels of protection,

exchange rates, import penetration and export shares between 1988 and 1995 accounts

for more than half of the observed changes in three out of four measures of inequality in

hourly wages.


The other key channel through which trade reform is likely to have affected the

distribution of wages is through changes in the economy-wide skill premium. This is the

channel on which Gonzaga et al. (2006) focused, and they argued that changes in the skill

premium mandated by a Stolper-Samuelson model of trade would account for more than

the actual change in skill-premium in manufacturing during 1988-1995. While our

approach is unable to identify changes in the economy-wide skill premium which are due

to trade variables from those which are not, our findings are consistent with the Gonzaga

et al. results: returns to education fell over the period, contributing to a decline in

inequality which did overshoot the observed decline. If there was any skill-biased

technical change, or if other forces for greater demand for skill were at work, they were

more than offset by the joint force of the Stolper-Samuelson effect of trade liberalization

in an economy that used to protect skill-intensive industries, and of increases in the

supply of more educated workers.


Overall, even if one does not attribute the decline in the economy-wide returns to

education to the trade reforms (despite evidence from other sources that part of it is

                                            33


attributable to the Stolper Samuelson effect), our results suggest that trade liberalization

did play an important part in the reduction of wage inequality in Brazil during 1988-

1995. The counterfactual wage growth incidence curve that includes the combined wage

and employment effects mandated by changes in trade variables accounted for 59% of the

observed change in the Theil-L index; 61% of the change in the Gini coefficient; and

76% of the change in the Theil-T index. Among the combined effects, changes in

occupation and employment levels across industries were by far the most important.

These reductions in wage inequality did appear to extend to declines in household income

inequality, and in the poverty rate.


Some of the implications of these findings are as follows. There is no reason why

researchers concerned with the distributional effects of trade liberalization should focus

exclusively on the manufacturing sector; or only on industry-specific wage premia.

Indeed, it would seem that employment flows and changes in the occupational structure

of the labor force play a central role, and should be considered explicitly. For policy-

makers, it would seem that in countries where protection was stronger for industries

intensive in skilled workers (which was not the case in Mexico, Chile or Colombia, but

was the case in Brazil), there need be no mandatory trade-off between gains in efficiency

and productivity on the one hand, and increases in inequality or poverty on the other.

Quite the contrary: the same liberalization efforts that lead to productivity gains may also

lead to wage gains at the bottom of the distribution, and to reductions in poverty and

inequality.

                                           34


References


Abreu, Marcelo de Paiva. 2004. "Trade Liberalization and the Political Economy of
   Protection in Brazil since 1987", INTAL-ITD Working Paper ­SITI- 08B, Inter-
   American Development Bank.

Acemoglu, Daron. 2003. "Patterns of Skill Premia", Review of Economic Studies 70: 199-
   230.

Attanasio, Orazio, Pinelopi Goldberg and Nina Pavcnik. 2004. "Trade Reforms and
   Income Inequality in Colombia", Journal of Development Economics 74(2): 331-336.

Beyer, Harold, Patricio Rojas, and Rodrigo Vergara. 1999. "Trade Liberalization and
   Wage Inequality." Journal of Development Economics 59: 103-123.

Bourguignon, Francois, Francisco Ferreira and Nora Lustig. 2004. The Microeconomics
   of Income Distribution Dynamics in East Asia and Latin America, Washington DC:
   The World Bank and Oxford University Press.

Cragg, Michael and Mario Epelbaum. 1996. "Why has Wage Dispersion Grown in
   Mexico? Is it the incidence of reforms or the growing demand for skills?" Journal of
   Development Economics 51: 99-116.

Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric
   Approach to Development Policy, Johns Hopkins University Press.

DiNardo, John, Nicole Fortin and Thomas Lemieux. 1996. "Labor Market Institutions
   and the Distribution of Wages, 1973-1992: A Semiparametric Approach",
   Econometrica 64(5): 1001-1044.

Ebrill Liam, Janet Stotsky, and Reint Gropp. 1999. "Revenue Implications of Trade
   Liberalization", IMF Occasional Paper 180, Washington D.C.

Edwards, Sebastian. 1998. "Openness, Productivity and Growth: What Do We Really
   Know?", Economic Journal 108: 383-398.

Feliciano, Zadia. 2001. "Workers and Trade Liberalization: The Impact of Trade Reforms
   in Mexico on Wages and Employment," Industrial and Labor Relations Review 55
   (1): 95-115.

Ferreira, Francisco H.G. and Ricardo Paes de Barros. 1999. "The Slippery Slope:
   Explaining the Increase in Extreme Poverty in Urban Brazil, 1976-1996." Brazilian
   Review of Econometrics 19 (2): 211-296.

Ferreira, Francisco H.G., Phillippe G. Leite, and Julie A. Litchfield. 2006. "The Rise and
   Fall of Brazilian Inequality: 1981-2004." World Bank Policy Research Working
   Paper #3867.

                                            35


Ferreira, Pedro Cavalcanti and José Luiz Rossi. 2003. "New Evidence from Brazil on
   Trade Liberalization and Productivity Growth". International Economic Review 44
   (4): 1383-1405.

Goldberg, Linda. 2004. "Industry-specific Exchange Rates for the United States", Federal
   Reserve Bank of New York Economic Policy Review, May 2004.

Goldberg, Pinelopi and Nina Pavcnik. 2004. "Trade, Inequality, and Poverty: What Do
   We Know? Evidence from recent trade liberalization episodes in developing
   countries", Brookings Trade Forum 2004, Brookings Institute.

Goldberg, Pinelopi and Nina Pavcnik. 2005. "Trade Protection and Wages: Evidence
   from Colombian Trade Reforms", Journal of International Economics, May 2005:
   75-105.

Gonzaga, Gustavo, Naércio Menezes-Filho and Cristina Terra. 2006. "Trade
   Liberalization and the Evolution of Skill Earnings Differentials in Brazil", Journal of
   International Economics 68(2): 345-367.

Hanson, Gordon and Ann Harrison. 1999. "Who Gains from Trade Reform? Some
   Remaining Puzzles", Journal of Development Economics 59(1): 125-154.

Juhn, Chinhui, Kevin Murphy and Brooks Pierce. 1993. "Wage Inequality and the Rise in
   Returns to Skill," Journal of Political Economy 101 (3): 410-42.

Kume, Honório, Guida Piani and Carlos Frederico de Souza. 2000. "A política brasileira
   de importaçăo no período 1987-98: descriçăo e avaliaçăo". IPEA, Rio de Janeiro,
   unpublished.

Lemieux, Thomas. 2002. "Decomposing Changes in Wage Distributions: a unified
   approach". Canadian Journal of Economics 35 (4): 646-688.

Muendler, Marc-Andreas. 2003. "Series of Market Penetration by Foreign Products,
   Brazil 1986-1998", University of California, San Diego.

Nissanke, Machiko and Erik Thorbecke. 2006. "Channels and Policy Debate in the
   Globalization-Inequality-Poverty Nexus". Cornell University, unpublished.

Pavcnik, Nina., Andreas Blom, Pinelopi Goldberg, and Norbert Schady. 2004. "Trade
   Policy and Industry Wage Structure: Evidence from Brazil", World Bank Economic
   Review 18(3): 319-344.

Ravallion, Martin and Shaohua Chen. 2003. "Measuring Pro-Poor Growth", Economics
   Letters 78: 93-99.

Ravallion, Martin and Michael Lokshin. 2004. "Winners and Losers from Trade Reform
   in Morocco", World Bank, Development Research Group, unpublished.

                                         36


Revenga, Ana. 1997. "Employment and Wage Effects of Trade Liberalization: the Case
   of Mexican Manufacturing", Journal of Labor Economics 15(3ii): S20-S43.

Robbins, Donald and T.H. Grindling. 1999. "Trade Liberalization and the Relative
   Wages for More-Skilled Workers in Costa Rica", Review of Development Economics
   3(2): 140-154.

Rodriguez, Francisco and Dani Rodrik. 2001. "Trade Policy and Economic Growth: A
   Skeptic's Guide to the Cross-National Evidence," in Ben Bernanke and Kenneth S.
   Rogoff, eds., NBER Macroeconomics Annual 2000, MIT Press for NBER.

Sachs, Jeffrey. and Andrew Warner. 1995. "Economic reform and the process of global
   integration.", Brookings Papers on Economic Activity, 1-118.

Shigeoka, Hitoshi, Eric Verhoogen and Matthew Wai-Poi. 2006. "Chinese Competition
   and the Mexican Manufacturing Sector: Plant-Level Evidence and Implications for
   Policy", Columbia University, unpublished.

Slaughter, Matthew. 2000. "What are the Results of Product-Prices Studies and What
   Can We Learn from their Differences?" In Feenstra, R. (ed.) The Impact of
   International Trade on Wages, NBER Conference Report, University of Chicago
   Press, pp.129-170.

Thoenig, Mathias and Thierry Verdier. 2003. "A Theory of Defensive Skill-Biased
   Innovation and Globalization", American Economic Review 93 (3): 709-728.

Verhoogen, Eric. 2006. "Trade Quality Upgrading and Wage Inequality in the Mexican
   Manufacturing Sector", Columbia University, unpublished.

Winters, Alan, Neil McCulloch and Andrew McKay. 2004. "Trade Liberalization and
   Poverty: the Evidence So Far", Journal of Economic Literature 62: 72-115.

Wood, Adrian. (1997) "Openness and Wage Inequality in Developing Countries ­ the
   Latin American Challenge to East Asian Conventional Wisdom", World Bank
   Economic Review 11 (1): 33-57.

                                                                                   37


Table 1: Trade Variables by Industry, 1988 and 1995

                                                                          Nom. Tariffs          ERP          Imp. Pen.     Exp. Share           M-RER            X-RER
Industry                                                                   88       95        88     95      88    95       88        95       88      95      88       95
Agricultural products                                                      17         7       15     8       NA    NA      NA         NA    5,347 7,351       110      193
Mining products                                                            20         3       15     0        8    12       98        77     149       196    135      209
Oil and coal extraction                                                     6         0       -3     -2      130   117       0         1       16      24     441      777
Non-metallic minerals                                                      39       10        46     12       1      3       2         4     105       169    180      297
Steel, non-ferrous and other metallurgy prdts.                             33       11        37     14       3      7      28        28       92      131   6,021 8,158
Machinery and tractors                                                     47       19        50     18      14    31       18        36     105       172    232      324
Electrical equipment, electronic equipment                                 50       21        59     30      12    18        7        10       56      86     114      182
Auto., trucks and buses; parts, comp. and other vehicles                   43       31        45     74       9    16       13        12       91      147    296      414
Wood products and furniture                                                30       11        29     12       0      1       4        10     311       540     29       45
Cellulose, paper and printing                                              32       10        30     10       2      6       6        11       40      55     143      221
Rubber products                                                            49       13        59     15       4    10        6         8     195       299    248      414
Chemical elements and products                                             66       15        76     16      56    87       33        39       30      46     118      189
Oil refining and petrochemicals                                            34         4       70     3       12    30       31         7     103       392     96      160
Pharmaceutical and perfumery products                                      45         8       52     8        6    13        2         3       45      73     208      346
Plastic products                                                           57       15        72     21       2      7       2         3       60      92     311      435
Textile products                                                           57       15        84     22      10    59       27        29     217       334    164      264
Apparel                                                                    76       20        94     24       0      3       1         2     162       253    109      185
Footwear                                                                   41       18        40     24       3      7      25        36     479       708     18       31
Processing of vegetal products                                             42       12        86     16       9    15       21        15     112       146     65      111
Meat packing, dairy industry, vegetal and other food products              63       20        73     21       3      7      50        47       52      83      85      140
Unclassified manufacturing                                                 49       14        64     15       0      0       0         0     101       169    102      265
Simple average                                                             44       17        52     21      19    28       22        22     390       566    456      652
Nontradables                                                                0         0       0      0        0      0       0         0       67      104     91      147
Sources: Nominal tariffs and effective rates of protection from Kume et al. (2000), reported in Abreu (2004); import penetration
and export share of production from Muendler (2003); import- and export-weighted real exchange rates are authors' calculations
from World Development Indicators (World Bank, 2006) and COMTRADE (UN, 2006). For more details, see Data Appendix.




Table 2: Industry Concordance

Trade Industry (Kume et al)             PNAD Code PNAD Industry                                                 Final Code Final Industry
Agricultural products                     11-42      Various crops, horticulture and forestry                       1     Agricultural products
Mining products                         50, 53-59    Prospecting and extraction of non-oil/gas/coal minerals        2     Mining products
Oil and coal extraction                   51-52      Oil, gas and coal                                              3      Oil and coal extraction
Non-metallic minerals                      100       Non-metal processing                                           4      Non-metallic minerals
Steel products                             110       Steel products                                                 5      Steel, non-ferrous and other metal products
Non-ferrous metallurgy                     110       Non-steel metals products                                      5      Steel, non-ferrous and other metal products
Other metallurgical products               110                                                                      5      Steel, non-ferrous and other metal products
Machinery and tractors                     120       Manufacture of machines and equipment                          6     Machinery and tractors
Electrical equipment                       130       Manufacture of electrical and electronic equipment             7      Electrical and electronic equipment
Electronic equipment                       130       Manufacture of electrical and electronic equipment             7      Electrical and electronic equipment
Automobiles, trucks and buses              140       Manufacture of vehicles and parts                              8     Automobiles, trucks and buses; parts, comp. and other vehicles
Parts, components and other vehicles       140       Manufacture of vehicles and parts                              8      Automobiles, trucks and buses; parts, comp. and other vehicles
Wood products and furniture           150, 151, 160 Manufacture of wood products and furniture                      9     Wood products and furniture
Cellulose, paper and printing            170, 290    Pulp and paper products, printing and newspapers              10     Cellulose, paper and printing
Rubber products                            180       Rubber products                                               11      Rubber products
Chemical elements                          200       Chemical products                                             12      Chemical elements and products
Oil refining                               201       Oil and petroleum products                                    13      Oil refining and petrochemicals
Chemical products                          200       Chemical products                                             12      Chemical elements and products
Pharmaceutical and perfumery products    210, 220    Pharmaceuticals and toiletries                                14      Pharmaceutical and perfumery products
Plastic products                           230       Plastics                                                      16      Plastic products
Textile products                         240, 241    Textiles                                                      17      Textile products
Apparel                                    250       Apparel and clothing                                          18      Apparel
Footwear                                   251       Footwear                                                      19      Footwear
Coffee industry                                                                                                    21      Meat packing, dairy industry, vegetal and other food products
Processing of vegetal products             280       Tobacco and other vegetal processing                          20      Processing of vegetal products
Meat packing                               260       Food preparation                                              21     Meat packing, dairy industry, vegetal and other food products
Dairy industry                             260       Food preparation                                              21      Meat packing, dairy industry, vegetal and other food products
Sugar                                      17?       Sugar cane extraction?                                        21     Meat packing, dairy industry, vegetal and other food products
Vegetal products                           260                                                                     21      Meat packing, dairy industry, vegetal and other food products
Other food products                   260, 261, 271 Other foods and drinks                                         21     Meat packing, dairy industry, vegetal and other food products
Other industries                           300       Various scientific instruments                                99      Unclassified manufacturing
                                         340-903     Construction, services, retail, finance, government etc.      22      Nontradables

Omitted                                    190       Leather and skins
                                           202       Manufacture of synthetic materials (nylon etc)

                                                                                 38


Table 3a: First Stage Regression Results: Wages

                                           1987         1988       1989       1990       1992       1993      1995        1996        1997       1998      1999
Male                                       0.321         0.32      0.336      0.297      0.249      0.271    0.222        0.214       0.228      0.214     0.216
                                         (0.007)*** (0.007)*** (0.007)*** (0.007)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.005)***
Experience                                 0.048        0.047      0.045      0.044      0.038      0.039    0.036        0.036       0.036      0.034     0.035
                                         (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)***
Experience squared                         -0.001       -0.001    -0.001     -0.001        0          0         0           0           0          0         0
                                         (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
White                                      0.135        0.143      0.185      0.157      0.139      0.153    0.156        0.165       0.165      0.159     0.161
                                         (0.005)*** (0.005)*** (0.006)*** (0.005)*** (0.005)*** (0.005)*** (0.005)*** (0.005)*** (0.005)*** (0.005)*** (0.004)***
Yellow                                     0.294        0.318      0.319      0.303      0.264      0.348    0.346        0.398       0.439      0.351     0.285
                                         (0.035)*** (0.038)*** (0.044)*** (0.038)*** (0.042)*** (0.040)*** (0.040)*** (0.043)*** (0.046)*** (0.035)*** (0.040)***
1-3 years education                        0.213         0.19      0.203      0.195      0.143      0.172    0.145        0.146        0.14      0.144     0.124
                                         (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.008)*** (0.009)*** (0.009)*** (0.008)*** (0.008)***
4 years education                          0.393         0.4       0.398      0.392      0.324      0.356     0.307       0.304       0.288      0.281     0.266
                                         (0.009)*** (0.010)*** (0.010)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)***
5-7 years education                        0.578        0.582      0.573      0.561      0.463      0.496    0.436        0.406        0.41      0.398     0.373
                                         (0.011)*** (0.011)*** (0.011)*** (0.011)*** (0.010)*** (0.010)*** (0.009)*** (0.010)*** (0.009)*** (0.009)*** (0.009)***
Completed primary                          0.815        0.812      0.806      0.767      0.666      0.709    0.628        0.605       0.602      0.586     0.547
                                         (0.012)*** (0.013)*** (0.013)*** (0.012)*** (0.011)*** (0.012)*** (0.011)*** (0.011)*** (0.011)*** (0.010)*** (0.010)***
9-10 years education                       0.955        0.967      0.981      0.944       0.82      0.848    0.746        0.741       0.726      0.705     0.664
                                         (0.015)*** (0.015)*** (0.016)*** (0.015)*** (0.014)*** (0.014)*** (0.013)*** (0.013)*** (0.012)*** (0.012)*** (0.012)***
Completed high school                       1.23        1.274      1.224      1.096      0.933      1.047    0.968        1.002       1.001      0.938     0.896
                                         (0.058)*** (0.060)*** (0.061)*** (0.049)*** (0.047)*** (0.049)*** (0.047)*** (0.047)*** (0.047)*** (0.046)*** (0.041)***
12-14 years education                      1.623        1.691      1.656      1.511      1.283      1.426     1.39        1.439       1.404       1.35     1.307
                                         (0.060)*** (0.062)*** (0.063)*** (0.051)*** (0.049)*** (0.052)*** (0.049)*** (0.049)*** (0.049)*** (0.048)*** (0.043)***
Completed university                        2.05        2.147       2.05      1.915      1.649       1.82    1.791        1.824       1.814      1.761     1.738
                                         (0.059)*** (0.061)*** (0.062)*** (0.050)*** (0.049)*** (0.051)*** (0.048)*** (0.049)*** (0.048)*** (0.047)*** (0.042)***
Formal employee                             0.22        0.333      0.237       0.17      0.401      0.357     0.184       0.182       0.198      0.207     0.228
                                         (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.005)*** (0.006)*** (0.005)*** (0.005)*** (0.005)***
Self-employed                              0.274        0.295      0.338      0.319      0.302      0.337    0.257        0.268       0.205      0.184     0.197
                                         (0.008)*** (0.008)*** (0.008)*** (0.008)*** (0.007)*** (0.008)*** (0.007)*** (0.007)*** (0.007)*** (0.007)*** (0.006)***
Employer                                   0.966        0.974       1.07      0.938      0.925      0.972    0.938        0.903       0.908      0.853     0.876
                                         (0.018)*** (0.019)*** (0.017)*** (0.015)*** (0.016)*** (0.016)*** (0.014)*** (0.016)*** (0.014)*** (0.014)*** (0.014)***
Northeast region                           -0.369       -0.353    -0.441     -0.443     -0.291     -0.404     -0.29       -0.276     -0.296     -0.228     -0.244
                                         (0.009)*** (0.010)*** (0.010)*** (0.010)*** (0.011)*** (0.011)*** (0.010)*** (0.010)*** (0.010)*** (0.010)*** (0.009)***
Southeast region                           -0.077       -0.025    -0.092     -0.124      0.097     -0.021    0.072        0.105        0.1        0.13     0.112
                                         (0.009)*** (0.009)*** (0.010)*** (0.009)*** (0.010)*** (0.010)** (0.010)*** (0.010)*** (0.009)*** (0.009)*** (0.009)***
South region                               -0.146       -0.107     -0.16     -0.165      0.063      0.026    -0.003       0.013       0.034      0.055      0.02
                                         (0.010)*** (0.011)*** (0.011)*** (0.011)*** (0.011)*** (0.011)**     -0.01       -0.011    (0.010)*** (0.010)*** (0.010)**
Central West region                        -0.019       -0.005    -0.079     -0.045      0.055      0.068    0.014        0.055       0.047      0.064     0.042
                                          (0.010)*      -0.01    (0.011)*** (0.010)*** (0.011)*** (0.011)***  -0.01    (0.011)*** (0.010)*** (0.010)*** (0.010)***
Metropolitan residence                     0.319        0.331      0.328      0.313      0.302      0.307    0.321        0.326       0.317      0.295     0.252
                                         (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.008)*** (0.009)*** (0.008)*** (0.008)*** (0.008)***
Urban residence                            0.198        0.127      0.146      0.158      0.135      0.145    0.157         0.15       0.122      0.108     0.091
                                         (0.008)*** (0.009)*** (0.009)*** (0.008)*** (0.009)*** (0.009)*** (0.008)*** (0.008)*** (0.008)*** (0.007)*** (0.007)***
Head of household                          0.208        0.203      0.193      0.189      0.159      0.166    0.175         0.17       0.156      0.156     0.149
                                         (0.007)*** (0.007)*** (0.007)*** (0.007)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.005)*** (0.005)*** (0.005)***

Industry indicators                         Yes          Yes        Yes        Yes        Yes        Yes      Yes          Yes         Yes        Yes       Yes
Industry*Skill indicators                   Yes          Yes        Yes        Yes        Yes        Yes      Yes          Yes         Yes        Yes       Yes

Observations                              112,655      112,730    114,961    116,882    118,075    119,949   128,360    124,017      131,202    129,719   133,310
R-squared                                   0.59         0.59       0.57       0.58       0.55       0.55     0.56         0.54        0.56       0.56      0.56

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Dependent variable is real hourly wages from principal job.
Regions are relative to North region. White and yellow are relative to Black. Educational attainment indicators are relative to no education.
Informal employee and self-employed coefficients are relative to formal employees.

Table 3b: First Stage Regression Results: Industry Wage Premiums

Industry                                                      1987    1988    1989    1990    1992    1993    1995    1996    1997    1998    1999
Mining products                                               0.514   0.578   0.488   0.433   0.238   0.315   0.316   0.213   0.347   0.393   0.349
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001)
Oil and coal extraction                                       0.678   0.666   0.680   0.727   0.835   0.874   0.817   1.043   0.677   0.824   0.994
                                                             (0.009) (0.011) (0.010) (0.011) (0.013) (0.018) (0.012) (0.018) (0.019) (0.013) (0.018)
Non-metallic minerals                                         0.202   0.203   0.225   0.242   0.188   0.191   0.274   0.269   0.353   0.353   0.314
                                                             (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Steel, non-ferrous and other metallurgy prdts.                0.398   0.398   0.373   0.458   0.381   0.378   0.476   0.484   0.494   0.492   0.432
                                                             (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Machinery and tractors                                        0.538   0.539   0.520   0.636   0.406   0.473   0.516   0.508   0.562   0.504   0.495
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Electrical equipment, electronic equipment                   0.467    0.493   0.450   0.614   0.507   0.404   0.551   0.503   0.611   0.577   0.458
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Auto., trucks and buses; parts, comp. and other vehicles      0.509   0.608   0.529   0.680   0.646   0.590   0.685   0.692   0.662   0.657   0.628
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Wood products and furniture                                   0.035   0.020   0.041   0.157   0.091   0.131   0.274   0.299   0.290   0.303   0.298
                                                             (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Cellulose, paper and printing                                 0.325   0.382   0.328   0.437   0.410   0.343   0.484   0.456   0.512   0.490   0.466
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Rubber products                                               0.480   0.536   0.325   0.515   0.293   0.387   0.553   0.665   0.433   0.394   0.473
                                                             (0.003) (0.003) (0.004) (0.003) (0.004) (0.004) (0.003) (0.005) (0.002) (0.005) (0.005)
Chemical elements and products                                0.410   0.326   0.300   0.394   0.383   0.288   0.421   0.421   0.458   0.403   0.454
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Oil refining and petrochemicals                              0.563    0.709   0.600   0.571   0.696   0.677   0.532   0.508   0.566   0.511   0.554
                                                             (0.010) (0.008) (0.011) (0.007) (0.007) (0.009) (0.008) (0.007) (0.004) (0.005) (0.011)
Pharmaceutical and perfumery products                         0.343   0.418   0.294   0.396   0.431   0.445   0.434   0.485   0.565   0.533   0.521
                                                             (0.003) (0.003) (0.004) (0.003) (0.003) (0.004) (0.004) (0.003) (0.004) (0.003) (0.003)
Plastic products                                              0.360   0.364   0.382   0.465   0.315   0.317   0.423   0.375   0.425   0.435   0.374
                                                             (0.001) (0.002) (0.003) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Textile products                                              0.153   0.239   0.207   0.253   0.195   0.223   0.321   0.285   0.386   0.338   0.184
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Apparel                                                       0.267   0.310   0.334   0.482   0.267   0.259   0.331   0.317   0.380   0.342   0.310
                                                             (0.000) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Footwear                                                      0.210   0.121   0.278   0.412   0.253   0.176   0.249   0.268   0.271   0.209   0.237
                                                             (0.000) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.000)
Processing of vegetal products                               0.306    0.360   0.192   0.276   0.522   0.408   0.640   0.477   0.426   0.446   0.174
                                                             (0.013) (0.008) (0.014) (0.008) (0.008) (0.011) (0.007) (0.010) (0.005) (0.014) (0.006)
Meat packing, dairy industry, vegetal and other food products 0.151   0.169   0.166   0.270   0.180   0.225   0.298   0.316   0.328   0.331   0.277
                                                             (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Unclassified manufacturing                                    0.068   0.065   0.100   0.160   0.080   0.071   0.095   0.196   0.159   0.147   0.103
                                                             (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000)
Nontradables                                                  0.164   0.149   0.149   0.300   0.184   0.171   0.306   0.328   0.371   0.367   0.327
                                                             (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Standard errors are in brackets.

Table 3c: First Stage Regression Results: Industry Skill Premiums

Industry                                                      1987    1988    1989    1990    1992    1993    1995    1996    1997    1998    1999
Mining products                                               0.089   0.108   0.267   0.400   0.462  -0.052   0.163   0.258   0.180   0.326   0.220
                                                             (0.018) (0.013) (0.010) (0.014) (0.015) (0.018) (0.014) (0.017) (0.013) (0.011) (0.012)
Oil and coal extraction                                       0.465   0.411   0.537   0.495   0.293   0.202   0.151   0.039   0.380   0.143   0.081
                                                             (0.020) (0.022) (0.021) (0.029) (0.023) (0.030) (0.022) (0.031) (0.032) (0.028) (0.025)
Non-metallic minerals                                         0.123   0.184   0.162   0.090   0.144   0.130   0.225   0.184   0.059   0.044   0.107
                                                             (0.008) (0.011) (0.013) (0.009) (0.009) (0.008) (0.006) (0.007) (0.007) (0.005) (0.005)
Steel, non-ferrous and other metallurgy prdts.                0.030  -0.071   0.096   0.125   0.148   0.132   0.090  -0.020   0.000   0.037   0.051
                                                             (0.005) (0.005) (0.006) (0.004) (0.004) (0.005) (0.004) (0.004) (0.004) (0.003) (0.003)
Machinery and tractors                                       -0.080  -0.119  -0.121   0.020   0.218   0.082   0.050  -0.008  -0.014   0.036   0.001
                                                             (0.006) (0.008) (0.008) (0.006) (0.005) (0.007) (0.006) (0.005) (0.004) (0.004) (0.004)
Electrical equipment, electronic equipment                    0.082   0.073   0.069   0.113   0.039   0.104   0.093  -0.035  -0.052   0.031   0.096
                                                             (0.006) (0.007) (0.007) (0.005) (0.007) (0.006) (0.005) (0.006) (0.005) (0.005) (0.004)
Auto., trucks and buses; parts, comp. and other vehicles     -0.007  -0.018  -0.072  -0.014   0.113   0.253   0.020  -0.104  -0.010   0.056   0.001
                                                             (0.006) (0.007) (0.007) (0.005) (0.006) (0.006) (0.004) (0.005) (0.004) (0.004) (0.004)
Wood products and furniture                                  -0.051   0.036  -0.125   0.121   0.050   0.020  -0.008  -0.179  -0.218  -0.059  -0.143
                                                             (0.010) (0.009) (0.009) (0.007) (0.006) (0.007) (0.007) (0.006) (0.006) (0.005) (0.003)
Cellulose, paper and printing                                -0.046  -0.131   0.000   0.046   0.006  -0.079   0.032  -0.069  -0.067   0.005  -0.067
                                                             (0.006) (0.008) (0.007) (0.006) (0.005) (0.005) (0.005) (0.005) (0.004) (0.004) (0.004)
Rubber products                                              -0.025  -0.273   0.054  -0.136   0.109   0.337   0.158  -0.018   0.179   0.258  -0.197
                                                             (0.032) (0.029) (0.022) (0.026) (0.025) (0.015) (0.017) (0.021) (0.024) (0.015) (0.021)
Chemical elements and products                                0.155   0.175   0.223   0.282   0.254   0.327   0.225   0.173   0.139   0.156   0.073
                                                             (0.007) (0.008) (0.007) (0.006) (0.006) (0.007) (0.005) (0.007) (0.006) (0.008) (0.005)
Oil refining and petrochemicals                               0.456   0.216   0.074   0.374   0.287   0.201   0.294   0.215   0.380   0.308   0.250
                                                             (0.019) (0.015) (0.026) (0.015) (0.017) (0.016) (0.017) (0.017) (0.013) (0.012) (0.018)
Pharmaceutical and perfumery products                         0.088  -0.057   0.184   0.211   0.193   0.078   0.243   0.152  -0.052   0.111   0.066
                                                             (0.011) (0.012) (0.015) (0.012) (0.011) (0.012) (0.011) (0.010) (0.011) (0.008) (0.007)
Plastic products                                              0.050   0.110  -0.080   0.170   0.125  -0.084   0.133   0.074   0.055  -0.030   0.043
                                                             (0.018) (0.014) (0.022) (0.010) (0.010) (0.011) (0.011) (0.010) (0.010) (0.008) (0.006)
Textile products                                             -0.002   0.153   0.086   0.117   0.235   0.145   0.181   0.143  -0.099   0.002   0.123
                                                             (0.010) (0.009) (0.009) (0.007) (0.007) (0.008) (0.007) (0.009) (0.007) (0.008) (0.006)
Apparel                                                      -0.185  -0.288  -0.224  -0.116  -0.160  -0.242  -0.230  -0.289  -0.119  -0.194  -0.135
                                                             (0.008) (0.008) (0.009) (0.006) (0.006) (0.005) (0.005) (0.005) (0.006) (0.005) (0.003)
Footwear                                                     -0.058  -0.062  -0.024   0.061  -0.105  -0.113  -0.155  -0.084  -0.157  -0.054  -0.153
                                                             (0.019) (0.015) (0.018) (0.009) (0.010) (0.010) (0.010) (0.012) (0.009) (0.007) (0.006)
Processing of vegetal products                                0.126   0.024  -0.069   0.329   0.242   0.001  -0.280   0.117   0.229   0.257   0.332
                                                             (0.025) (0.041) (0.024) (0.031) (0.023) (0.022) (0.047) (0.022) (0.032) (0.032) (0.017)
Meat packing, dairy industry, vegetal and other food products 0.000  -0.009   0.019   0.010   0.125  -0.003   0.069  -0.064  -0.006   0.016   0.032
                                                             (0.005) (0.006) (0.006) (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.002)
Unclassified manufacturing                                    0.110   0.056   0.282   0.357   0.233   0.035   0.232   0.081   0.140   0.137   0.167
                                                             (0.008) (0.012) (0.010) (0.008) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.004)
Nontradables                                                  0.194   0.141   0.185   0.265   0.204   0.158   0.130   0.055   0.053   0.095   0.101
                                                             (0.003) (0.004) (0.004) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Standard errors are in brackets.

Table 4a: First Stage Regression Results, 1988 and 1995: Employment multinomial logit - Marginal Effects
                                                                                                                1988
                                          0               1                2               3               4               5              6            7                8                9
Probability (at mean)                0.329            0.026            0.186          0.046            0.013          0.016           0.080        0.108           0.173             0.022
Male                                                  0.009     ***    0.137    ***   0.091     ***    0.025    ***   0.017    ***    0.081    *** -0.010    ***   0.087     ***     0.032   ***
Experience                                            0.001     ***    0.005    ***   -0.001    ***    0.000    ***   -0.001   ***    0.008    *** -0.010    ***   0.015     ***     0.005   ***
Experience squared                                    0.000     ***    0.000    ***   0.000     ***    0.000    ***   0.000    ***    0.000    *** 0.000     ***   0.000     ***     0.000   ***
White                                                 -0.008    ***    0.013    ***   -0.015    ***    -0.002   ***   -0.001          0.005    *** -0.031    ***  -0.026     ***     0.015   ***
Yellow                                                -0.030    ***    0.093    *     0.042     **     -0.032   ***   0.003           0.006        -0.098    ***  -0.076     ***     0.022   ***
1-3 years education                                   0.007     ***   -0.013          -0.030    ***    -0.004   ***   -0.001          0.020    *** 0.007     ***   0.032     ***     0.013   ***
4 years education                                     0.005     ***   -0.010    ***   -0.056    ***    -0.008   ***   -0.002          0.039    *** -0.003    ***   0.065     ***     0.024   ***
5-7 years education                                   0.013     ***   -0.048    ***   -0.088    ***    -0.014   ***   -0.003          0.048    *** 0.003     ***   0.088     ***     0.029   ***
Completed primary                                     0.012     ***   -0.044    **    -0.120    ***    -0.018   ***   -0.006   **     0.051    *** -0.005          0.130     ***     0.040   ***
9-10 years education                                  0.009     ***   -0.071    ***   -0.130    ***    -0.026   ***   -0.009   ***    0.042    *** -0.011          0.165     ***     0.042   ***
Completed high school                                 0.018     ***   -0.074          -0.131    ***    -0.017   ***   -0.011   **     0.055    *** 0.033     ***   0.222     ***     0.052   ***
12-14 years education                                 0.001           -0.122    ***   -0.188    ***    -0.033   ***   -0.013   ***    0.046    *** 0.078     ***   0.229     ***     0.054   ***
Completed university                                  0.009     ***   -0.039    ***   -0.132    ***    -0.017   *     -0.012          0.053    *** 0.117     ***   0.238     ***     0.058   ***
Region 2                                              0.008     ***   -0.013          0.027     ***    0.014    ***   0.000           -0.014   **  -0.003         -0.002             -0.002
Region 3                                              0.005     ***   -0.085    ***   0.036     ***    0.030    ***   -0.005          0.064    *** -0.002    ***   0.024     ***     0.001   ***
Region 4                                              0.009     ***   -0.021    ***   0.011     ***    0.023    ***   -0.007          0.059    *** -0.021    ***   0.054     ***     -0.001  ***
Region 5                                              0.000           -0.052    ***   0.038     ***    0.021    ***   0.001           -0.048   *** 0.021     ***   0.044     ***     0.003   ***
Metropolitan residence                                0.033     ***   -0.203    ***   -0.163    ***    -0.041   ***   0.006    ***    0.099    *** 0.073     ***   0.191     ***     -0.013  ***
Urban residence                                       0.026     ***   -0.198    ***   -0.069    ***    -0.019   ***   0.004    ***    0.066    *** 0.082     ***   0.140     ***     -0.007  ***
Head of household                                     0.004     ***    0.095    ***   0.014     ***    0.014    ***   0.007    ***    0.082    *** 0.065     ***   0.145     ***     0.030   ***
Children 0-14 years in household                      -0.003    ***   -0.011    ***   -0.001    ***    0.001    ***   -0.001   ***    -0.014   *** -0.005    ***  -0.042     ***     -0.002  ***
Has a spouse who works                                -0.972    ***    1.870    ***   0.439     ***    0.124          0.159    ***    0.753    *** 1.033     ***   1.660     ***     0.218   ***

Observations                                                                                                   177,376
                                                                                                                1995
                                          0               1                2               3               4               5              6            7                8                9
Probability (at mean)                0.308            0.046            0.203          0.034            0.013          0.015           0.061        0.103           0.190             0.027
Male                                                  -0.004    ***    0.160    ***   0.073     ***    0.026    ***   0.015     ***   0.061    *** -0.040    ***   0.050     ***     0.036   ***
Experience                                            0.001     ***    0.008    ***   0.000     ***    0.001    ***   -0.001    ***   0.007    *** -0.005    ***   0.021     ***     0.006   ***
Experience squared                                    0.000     ***    0.000    ***   0.000     ***    0.000    ***   0.000     ***   0.000    *** 0.000     ***   0.000     ***     0.000   ***
White                                                 -0.008    ***    0.023          -0.013    ***    -0.002   ***   -0.002    ***   0.003     ** -0.022    ***  -0.028     ***     0.014   ***
Yellow                                                -0.029    ***    0.114    ***   0.016            -0.006         -0.008          -0.017    ** -0.041    ***  -0.133     ***     0.039   ***
1-3 years education                                   -0.001    ***    0.004    ***   -0.022    ***    -0.005   ***   -0.002          0.019    *** -0.004    ***   0.052     ***     0.010   ***
4 years education                                     -0.003    ***    0.001    ***   -0.039    ***    -0.009   ***   -0.001    ***   0.034    *** -0.017    ***   0.092     ***     0.025   ***
5-7 years education                                   0.006     ***   -0.016    ***   -0.062    ***    -0.013   ***   0.000     ***   0.043    *** -0.021          0.115     ***     0.029   ***
Completed primary                                     0.008     ***   -0.017    ***   -0.080    ***    -0.026   ***   -0.003          0.047    *** -0.049    ***   0.171     ***     0.038   ***
9-10 years education                                  0.008     ***   -0.042      *   -0.095    ***    -0.031   ***   -0.007          0.046    *** -0.045    ***   0.206     ***     0.049   ***
Completed high school                                 0.006     ***   -0.028    ***   -0.103    ***    -0.028   ***   -0.007          0.054    *** -0.053    ***   0.293     ***     0.059   ***
12-14 years education                                 -0.003    ***   -0.048    ***   -0.150    ***    -0.036   ***   -0.010          0.048    *** -0.045          0.321     ***     0.068   ***
Completed university                                  -0.009    ***   -0.016    ***   -0.096    ***    -0.027   ***   -0.010          0.059    *** -0.049    ***   0.358     ***     0.075   ***
Region 2                                              -0.011           0.016    ***   0.008     ***    0.018    ***   -0.005    ***   -0.007       0.006     ***   0.000      **     -0.002
Region 3                                              -0.012    ***   -0.081    ***   0.023     ***    0.035    ***   -0.004          0.054    *** 0.002     ***   0.042     ***     -0.001  ***
Region 4                                              -0.007    ***   -0.036    ***   0.014     ***    0.031    ***   -0.004          0.062    *** -0.003    ***   0.043     ***     0.004   ***
Region 5                                              -0.006    ***   -0.071    ***   0.029     ***    0.028    ***   -0.004    ***   -0.021   *** 0.020     ***   0.029     ***     0.003   *
Metropolitan residence                                0.054     ***   -0.216    ***   -0.108    ***    -0.049   ***   0.003           0.048    *** 0.074     ***   0.186     ***     -0.016  ***
Urban residence                                       0.045     ***   -0.216    ***   -0.047    ***    -0.019   ***   0.002     ***   0.033    *** 0.077     ***   0.139     ***     -0.011  ***
Head of household                                     0.011     ***    0.013    ***   0.007     ***    0.012    ***   0.003     ***   0.046    *** 0.040     ***   0.104     ***     0.022   ***
Children 0-14 years in household                      0.005     ***   -0.024    ***   0.000     ***    0.001     *    -0.001    ***   -0.003   *** -0.007    ***  -0.028     ***     -0.005  ***
Has a spouse who works                                -0.055    ***    0.156    ***   0.013     ***    0.005    ***   0.005     ***   0.019    *** 0.021     ***   0.077     ***     0.021   ***

Observations                                                                                                   208,400
Marginal effects reported. White and yellow are race indicators (black is omitted category); 0 years of education is omitted education category.
Columns headed 0-9 indicate ML results for each employment classification: 0. Not economically active; 1. Unemployed; 2. Self-employed (all industries); 3. Informal agriculture; 4. Formal agricu
5. Informal manufacturing; 6. Formal manufacturing; 7. Informal nontradable sector; 8. Formal nontradable sector; 9. Employer (all industries).

Table 4b: First Stage Regression Results, 1988 and 1995: Employment multinomial logit - industry participation constant

Industry                 1987        1988         1989         1990   1992     1993       1995       1996       1997       1998       1999
Unemployed              -4.428       -3.787      -4.476       -4.234  -2.984   -2.783    -2.978     -3.106     -2.802     -2.659      -2.572
                      (0.134)*** (0.124)*** (0.142)*** (0.130)*** (0.092)*** (0.092)*** (0.091)*** (0.087)*** (0.080)*** (0.077)*** (0.072)***
Self-employed           -1.157       -1.177      -1.052       -1.168  -1.427  -1.523      -1.75     -2.158     -2.106     -2.135      -2.054
                      (0.060)*** (0.060)*** (0.060)*** (0.059)*** (0.057)*** (0.057)*** (0.055)*** (0.055)*** (0.054)*** (0.054)*** (0.053)***
Informal agr.           -1.374       -1.973       -1.39       -1.483  -2.161   -2.253    -2.648     -3.175     -2.949     -2.631      -2.73
                      (0.117)*** (0.128)*** (0.125)*** (0.120)*** (0.128)*** (0.125)*** (0.129)*** (0.135)*** (0.128)*** (0.129)*** (0.129)***
Formal agr.             -4.487       -5.104       -4.55       -4.358  -5.934   -6.382    -5.984     -6.483      -7.11     -6.019      -6.919
                      (0.240)*** (0.276)*** (0.284)*** (0.236)*** (0.311)*** (0.314)*** (0.291)*** (0.310)*** (0.417)*** (0.294)*** (0.388)***
Informal manuf.         -3.161       -3.059      -3.054       -2.944  -3.73    -3.433    -3.614      -4.07     -4.038     -4.096      -4.408
                      (0.141)*** (0.140)*** (0.133)*** (0.139)*** (0.145)*** (0.135)*** (0.138)*** (0.141)*** (0.133)*** (0.140)*** (0.141)***
Formal manuf.           -4.395       -4.462      -4.116       -4.136  -5.138   -5.153    -5.394     -5.524     -5.273     -5.474      -5.302
                      (0.091)*** (0.092)*** (0.090)*** (0.091)*** (0.101)*** (0.101)*** (0.102)*** (0.101)*** (0.097)*** (0.102)*** (0.103)***
Informal nontrad.       -1.369       -1.578      -1.456        -1.44  -1.324   -1.206    -1.558     -1.753     -1.744     -1.636      -1.791
                      (0.064)*** (0.065)*** (0.065)*** (0.063)*** (0.066)*** (0.062)*** (0.062)*** (0.060)*** (0.059)*** (0.058)*** (0.058)***
Formal nontrad.         -3.123       -2.967      -2.841       -2.919  -3.62    -3.575    -3.802     -3.727     -3.797     -3.802      -3.814
                      (0.064)*** (0.063)*** (0.063)*** (0.063)*** (0.062)*** (0.062)*** (0.061)*** (0.058)*** (0.058)*** (0.059)*** (0.058)***
Employer                -7.571       -8.099      -6.581       -6.615  -7.724   -8.007     -7.83     -8.185     -8.477     -8.365      -8.205
                      (0.149)*** (0.156)*** (0.128)*** (0.122)*** (0.139)*** (0.145)*** (0.136)*** (0.144)*** (0.137)*** (0.138)*** (0.135)***
Observations           177,399     177,376      179,938      184,724 193,931  197,658    208,400    209,264    219,710    221,088    227,369
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Non-economically active is omitted category.




Table 5: Industry Wage Premiums and Trade Exposure

Dependent variable is industry wage premium                    (1)       (2)       (3)         (4)          (5)          (6)          (7)        (8)
ERP                                                          0.0005    0.0005    0.0005
                                                            (0.0005)  (0.0006)  (0.0006)
ERP * lagged import penetration                                       -0.0001    -0.0005
                                                                      (0.0019)  (0.0019)
Lagged ERP                                                                                   0.0003       0.0003       0.0005       0.0003     0.0005
                                                                                            (0.0001)*   (0.0001)** (0.0002)*** (0.0001)** (0.0002)***
Lagged ERP * lagged import penetration                                                                                -0.0015                 -0.0015
                                                                                                                      (0.0011)                (0.0012)
Lagged import penetration                                             -0.1531    0.0992                  -0.1196      -0.0926       0.1245     0.1508
                                                                      (0.1533)  (0.1757)                 (0.1452)     (0.1505)     (0.1645)   (0.1746)
Lagged export share                                                   -0.0524   -0.1251                  -0.0284      -0.0282      -0.1008    -0.1008
                                                                      (0.1313)  (0.1354)                 (0.1298)     (0.1306)     (0.1334)   (0.1341)
Lagged imp. pen * lagged imp.weighted RER                                         -0.001                                            -0.001     -0.001
                                                                               (0.0004)***                                       (0.0004)*** (0.0004)***
Lagged exp.shr * lagged exp.weighted RER                                         0.0001                                             0.0001     0.0001
                                                                               (0.0000)**                                         (0.0000)** (0.0000)**

Observations                                                   210      207        207         210          207         207          207        207
R-squared                                                      0.01     0.01       0.03        0.01        0.02         0.02         0.03       0.04
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

                                                                      43


Table 6: Industry Skill Premiums and Trade Exposure

Dependent variable is industry skill premium               (1)       (2)        (3)        (4)         (5)         (6)         (7)        (8)
ERP                                                      -0.0004      0          0
                                                        (0.0005)  (0.0007)   (0.0007)
ERP * lagged import penetration                                    -0.0031    -0.0027
                                                                  (0.0028)   (0.0027)
Lagged ERP                                                                               -0.0003    -0.0004     -0.0004     -0.0003    -0.0004
                                                                                        (0.0003)    (0.0003)    (0.0003)    (0.0003)   (0.0003)
Lagged ERP * lagged import penetration                                                                          -0.0001                 0.0002
                                                                                                                (0.0012)               (0.0014)
Lagged import penetration                                          -0.2762    -0.486                -0.3476     -0.3453     -0.3578     -0.361
                                                                  (0.1652)* (0.1855)***            (0.1509)** (0.1551)**   (0.1886)*  (0.1928)*
Lagged export share                                                0.0746     0.1151                 0.0445      0.0446      0.0826     0.0826
                                                                  (0.1493)   (0.1577)               (0.1510)    (0.1514)    (0.1591)   (0.1595)
Lagged imp. pen * lagged imp.weighted RER                                      0.001                                         0.0012     0.0012
                                                                             (0.0005)*                                     (0.0005)** (0.0005)**
Lagged exp.shr * lagged exp.weighted RER                                         0                                           0.0001     0.0001
                                                                             (0.0001)                                       (0.0001)   (0.0001)

Observations                                               204       201        201        208         201        201         204        204
R-squared                                                 0.00      0.01       0.02       0.01        0.02        0.02        0.02       0.02
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%




Table 7: Industry Participation and Trade Exposure

Dependent variable is industry participation               (1)       (2)        (3)        (4)         (5)         (6)         (7)        (8)
ERP                                                     0.0087    -0.0025    -0.0043
                                                        (0.0076)  (0.0213)   (0.0229)
ERP * lagged import penetration                                    0.0849     0.0965
                                                                  (0.1407)   (0.1574)
Lagged ERP                                                                              -0.0148     -0.0139      -0.015      -0.015    -0.0161
                                                                                       (0.0041)*** (0.0045)*** (0.0047)*** (0.0069)** (0.0077)**
Lagged ERP * lagged import penetration                                                                           0.0437                 0.0483
                                                                                                                (0.0326)               (0.0723)
Lagged import penetration                                         -1.5541    -0.2417                (1.3357)    (1.2278)    0.4760     1.2010
                                                                  (3.9782)   (5.7256)               (1.4342)    (1.5567)    (3.8118)   (4.6518)
Lagged export share                                                5.9193     4.9226                  3.328      4.0074      2.5479     2.9954
                                                                 (2.0965)*** (3.4338)              (1.1175)*** (1.0568)*** (2.3503)    (2.3160)
Lagged imp. pen * lagged imp.weighted RER                                    -0.0091                                        -0.0049    -0.0024
                                                                             (0.0152)                                       (0.0105)   (0.0080)
Lagged exp.shr * lagged exp.weighted RER                                      0.0053                                        -0.0014     0.005
                                                                             (0.0183)                                       (0.0152)   (0.0223)

Observations                                               80        72         36         72          72          72          36         36
R-squared                                                 0.02      0.02       0.05       0.05        0.06        0.06        0.11       0.12
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

                                                     44




    Table 8: Actual and Counterfactual Hourly Wage Distributions
                                                                  P90/P10 GE(0)    GE(1)    Gini
wij = exp Xij  88 + Iij * wp88 + (Iij * Sij )sp88 + F88 (i )
 88      (   88        88             88     88          -1     )
                               j                   j          88  16.9   0.703 0.780 0.611
wij = exp Xij  88 + Iij * wpsj + (Iij * Sij )sp88 + F88 (i )
 1       ( 88         88             88    88           -1     )
                                                  j         88    16.9   0.705 0.784 0.611
wij = exp Xij  88 + Iij * wpsj + (Iij * Sij )spsj + F88 (i )
 2       ( 88         88             88    88          -1     )
                                                            88    16.7   0.699 0.774 0.609
wij = exp Xij  88 + Iij * wpsj + (Iij * Sij )spsj + F88 (i )
 3       ( 88         s            s     88          -1      )
                                                          88      14.6   0.653 0.731 0.593
wij = exp Xij  + Iij * wp95 + (Iij * Sij )sp95 + F88 (i )
 4       (  88  s    s             s     88           -1      )
                           j                   j           88     12.9   0.600 0.669 0.572
wij = exp Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F88 (i )
 5       ( 88         s              s     88          -1     )
                            j                    j          88    12.3   0.581 0.657 0.566
wij = exp Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F95 (i )
 6       ( 88         s              s     88          -1     )
                            j                    j          88    12.0   0.587 0.691 0.571
wij = exp Xij  95 + Iij * wp95 + (Iij * Sij )sp95 + F95 (i )
 95      (   95        95              95    95           -1    )
                               j                   j          95  12.4   0.617 0.715 0.582

    Source: Author's Calculation from PNADs.




    Table 9: Actual and Counterfactual Household Per Capita Income
    Distributions
                                                                      Poverty Line R$87.55
               P90/P10     GE(0)         GE(1)         Gini
                                                                 FGT(0)     FGT(1)       FGT(2)
    HPCIij88     19.3      0.717         0.750        0.609       27.7        11.5        6.7

    HPCIij1      19.9      0.729         0.760        0.613       27.7        12.1        7.1

    HPCIij2      19.7      0.724         0.755        0.611       27.8        12.1        7.1

    HPCIij3      16.9      0.658         0.692        0.589       21.2        8.7         4.9

    HPCIij4      15.3      0.613         0.644        0.571       21.2        8.6         4.9

    HPCIij5      15.4      0.616         0.645        0.572       19.0        7.6         4.3

    HPCIij6      15.2      0.621         0.663        0.575       19.7        7.8         4.4

    HPCIij95     16.9      0.660         0.706        0.592       23.8        10.5        6.3

    Note: This is a relative poverty line, calculated to represent 50% of the
    median household per capita income in 1988 (expressed in 2004 prices.
    Source: Authors' calculations from the PNAD.

                                                                                       45



                                            Figure 1: Protection and Import Penetration in Brazil, 1985-1999


                                     60                                                                                                  35


                                     50                                                                                                  30

             )                                                                                                                                 )ega
              esgatn                                                                                                                     25
                                     40

                    cerep(                                                                                                               20        ntecrep(

                                     30
                          noit                                                                                                           15                onitart

                              ectorP 20                                                                                                  10                       neePt
                                                                                                                                                                       por
                                     10                                                                                                  5                                Im


                                      0                                                                                                  0

                                         1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

                                                                                      Year


                                                    effective rate of protection        nominal tariffs       import penetration




Source: Tariffs and rates of protection from Kume et al. (2000) presented in Abreu (2004). Import penetration from Muendler (2003).




                                          Figure 2: Household Per Capita Income Inequality in Brazil, 1981-2004

            0.64                                                                                                                           0.88




            0.62                                                                                                                           0.84
                                                                      0.609


            0.60                                                                                   0.591                                   0.80




            0.58                                                      0.750                                                                0.76

   Index
        ni
          Gi0.56                                                                                   0.705                                   0.72                              xednIilehT




            0.54                                                                                                                           0.68




            0.52                                                                                                                           0.64




            0.50                                                                                                                           0.60
                                      1981 1982 1983  1984 1985 1986 1987 1988 1989 1990 1992 1993 1995 1996 1997 1998 1999 2001 2002 2004

                                                                                       Year


 Source: Ferreira, Leite and Litchfield, 2006.                                      Gini      Theil

                                                                  46


.


                                 Figure 3: Hourly Wage Inequality in Brazil, 1987-2004

            0.64                                                                                                     1.02




            0.62       0.611                                                                                         0.96




            0.60                                                                                                     0.90


                                                            0.582
            0.58                                                                                                     0.84

   Index
        ni            0.780
          Gi0.56                                                                                                     0.78 xednIilehT



                                                            0.715
            0.54                                                                                                     0.72




            0.52                                                                                                     0.66




            0.50                                                                                                     0.60
                  1987   1988      1989   1990  1992   1993   1995    1996    1997   1998   1999  2001   2002   2004

                                                                  Year


 Source: Authors' calculation from PNADs.                      Gini       Theil

                                                                                                 47




                                                                 Figure 4: Skill Wage Premium and Share of Skilled Workers in Total
                                                                                          Employment, 1987-2004
                               45%                                                                                                                   4.5


                               43%                                  4.06                                                                             4.3


                               40%                                                                                                                   4.0


                               38%                                                                                                                   3.8 de
                                                                                                  3.48
         s                                                                                                                                                 kill
                               35%                                                                                                                   3.5       Unot
          Worker               33%                                                                                                                   3.3           led                   ges
                ledilkS        30%                                                                                                                   3.0              kilS                  Wa

                                                                                                                                                                          age
                       ofearhS 28%                                                                                                                   2.8                                      Hourly

                               25%                                                                                                                   2.5                     Averfoo
                               23%                                                                23.7%                                              2.3                            Rati

                               20%                                                                                                                   2.0


                               18%                                  20.2%                                                                            1.8


                               15%                                                                                                                   1.5
                                                              1987  1988  1989  1990  1992  1993   1995    1996  1997  1998  1999  2001  2002  2004

                                                                                                       Year
      Source: Authors' calculation from PNADs.                                                                                         Share    Ratio
      Note: Unskilled workers have 10 or fewer years of schooling. Skilled workers have 11 or more years of schooling.




                                                                         Figure 5: Change in Employment versus Change in
                                                                               ERP*Imp.Pen. by Industry, 1987-1995


                                                                                                                                               0.30

                                  95                                                                                                           0.20
                                    19
                                                                                                                                               0.10
                                      1987-,
                                            nte                                                                                                0.00

                                               oym                                                                                             -0.10
                                                  pl
                                                    mE                                                                                         -0.20
                                                      ni
                                                                                                                                               -0.30
                                                        nge
                                                           haC                                                                                 -0.40

                                                                                                                                               -0.50
                                                                -1.50    -1.00      -0.50       0.00        0.50      1.00       1.50       2.00

                                                                                     Change in ERP*Imp.Pen, 1987-1995


Source: Authors' calculations. Rates of protection from Kume et al. (2000) presented in Abreu (2004); import penetration from

Muendler (2003); employment from PNADs.

                                                             48



                    Figure 6: Observed and counterfactual wage growth incidence curves,
                                            1995-1988: industry wage premia.

     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0              10           20     30      40          50          60       70    80    90    100

                                                              percentiles


Source: Authors' calculation from PNADs.                      g (p)       g 1(p)




                    Figure 7: Observed and counterfactual wage growth incidence curves,
                                         1995-1988: industry and skill wage premia.

     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0              10           20     30      40          50          60       70    80    90    100

                                                              percentiles


Source: Authors' calculation from PNADs.                g (p)       g 1(p)      g 2(p)

                                                            49



                    Figure 8: Observed and counterfactual wage growth incidence curves,
                                1995-1988: all trade-mandated changes from 2nd stage.

     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0              10           20     30     40        50         60       70    80    90    100

                                                             percentiles


Source: Authors' calculation from PNADs.               g (p)      g 2(p)    g 3(p)




                    Figure 9: Observed and counterfactual wage growth incidence curves,
                                          1995-1988: upper-bound on trade effects.

     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0              10           20     30     40        50         60       70    80    90    100

                                                             percentiles


Source: Authors' calculation from PNADs.               g (p)      g 3(p)    g 4(p)

                                                            50



                  Figure 10: Observed and counterfactual wage growth incidence curves,
                                      1995-1988: trade effects + other price changes.

     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0              10           20    30      40        50         60       70    80    90    100

                                                             percentiles


Source: Authors' calculation from PNADs.               g (p)      g 4(p)    g 5(p)




                  Figure 11: Observed and counterfactual wage growth incidence curves,
                           1995-1988: trade effects, price changes + changes in residuals.

     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0              10           20    30      40        50         60       70    80    90    100

                                                             percentiles


Source: Authors' calculation from PNADs.               g (p)      g 5(p)    g 6(p)

                                                                51



                     Figure 12: Observed and counterfactual household per capita income growth
                                      incidence curves, 1995-1988: industry wage premia.


     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20 25  30   35  40     45     50    55    60   65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                         g (p)       g 1(p)




                     Figure 13: Observed and counterfactual household per capita income growth
                                  incidence curves, 1995-1988: industry and skill wage premia.


     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20 25  30   35  40     45     50    55    60   65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                  g (p)        g 1(p)      g 2(p)

                                                                52




                     Figure 14: Observed and counterfactual household per capita income growth
                     incidence curves, 1995-1988: all trade-mandated changes from second stage.


     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20 25   30   35 40     45   50      55 60   65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                  g (p)       g 2(p)    g 3(p)




                     Figure 15: Observed and counterfactual household per capita income growth
                                   incidence curves, 1995-1988: upper-bound on trade effects.


     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20 25   30   35 40     45   50      55 60   65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                  g (p)       g 3(p)    g 4(p)

                                                                53



                     Figure 16: Observed and counterfactual household per capita income growth
                                incidence curves, 1995-1988: trade effects + other price changes.


     110%



      90%



      70%



      50%

   %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20 25  30   35  40     45   50      55 60    65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                  g (p)       g 4(p)    g 5(p)




                     Figure 17: Observed and counterfactual household per capita income growth
                                    incidence curves, 1995-1988: trade effects, price changes
                                                    + changes in residuals.
     110%



      90%



      70%



      50%

    %

      30%



      10%



     -10%



     -30%
            0       5      10     15    20  25 30   35  40     45   50      55 60    65  70 75 80 85 90 95 100

                                                                 percentiles


Source: Authors' calculation from PNADs.                  g (p)       g 5(p)     g 6(p)

                                                 54


Data Appendix


This data appendix outlines how each trade variable used in our second-stage regressions

was constructed. The main trade variables are industry-specific real exchange rates;

effective rates of protection; import penetration rates and export shares of production.


Industry-specific trade-weighted real exchange rates

Adapting Goldberg (2004), we construct both export- and import-weighted real exchange

rates for each industry j. Goldberg suggests, respectively:

xert = wt × rert , where wt =
    j        jc      c           jc     Xtjc


         c                            X         , and
                                             jc
                                           t
                                       c


mert = wt × rert , where wt =
     j        jc     c            jc    Mt  jc


          c                           M          ,
                                               jc
                                              t
                                        c


where rert are the bilateral real exchange rates with each Brazilian trading partner c.
           c


Rather than deriving real exchange rates for every trading partner by industry and year, a

large undertaking, we have ranked countries by decreasing size of imports (exports) in

each industry, and included those countries who (i) make up the first 95% of imports

(exports), and (ii) whose imports (exports) are over 1% of the total within that industry.

To reduce spurious volatility in the exchange rate series due to a changing composition of

the set of countries in the annual weights, we depart from Goldberg by defining our

weighted exchange rates with constant weights across the period:

                                           1999
                                           X        jc
                                                   t
xert = w8599 × rert , where w8599 =
    j        jc       c            jc     t=1985           , and
                                              1999
         c                                X            jc
                                                      t
                                         c t=1985

                                            1999
                                            M         jc
                                                     t
mert = w8599 × rert , where w8599 =
     j        jc       c            jc     t=1985           .
                                               1999
          c                               M              jc
                                                        t
                                          c t=1985

These weights are based on a country's share of trade over the period 1985-1999. All

trading partners of significance over the period are included, and their real exchange rates

                                            55


are weighted identically in each year, so the variation in the aggregate industry exchange

rate comes from changes in trading partners' exchange rates, rather than trade volume or

inclusion/exclusion in different years.


We use trade data from COMTRADE at the SITC 4-digit level. A concordance was

constructed to match each SITC2 4-digit industry code to the more aggregated industries

of the ERP data from Kume et al. Imports (exports) are then summed by country over

these ERP industries to give us Mt       jc ( Xt ). Bilateral real exchange rates were
                                                jc


constructed by multiplying each country's nominal exchange rate (in local currency per

Real) by the ratio of the Brazilian price index to the partner country price index, where

the WPI (wholesale price index) was used where available, and the CPI (consumer price

index) otherwise. Countries for which neither a full time series of WPI or CPI were

available were excluded. For the nontradables sector, we constructed economy-wide real

exchange rates, using imports and exports with trading partners across all industries to

derive country weights.


Effective rates of protection

ERPs from 1985-1999 are available from Kume et al. (2000). We have used them as

reported in Abreu (2004). However, the industry classifications differ from those

available in the PNAD. Adapting the concordance used by Pavcnik et al. (2004), we

averaged ERPs across certain ERP industries using lagged industry imports as weights.

The summary final concordance is reported in Table 2 in the main text.


Import penetration rate and export share of production

Both of these variables were constructed in the same way; the raw data for both is from

Muendler (2003), and available at http://www.econ.ucsd.edu/muendler/html/brazil.html.

These data are given by Nível 80 (an official Brazilian classification), a different industry

classification than that afforded us by the PNAD, so a concordance from Nível 80 to

Nível 100 (also available on Muendler's website) was used to move to Nível 100, and a

second concordance from Nível 100 to the ERP industry categories was constructed by

the authors to standardize industries. As with ERPs, import weights were then used to

                                                    56


average certain industry import penetration and export share to arrive at data series for

the final industries used in the paper.


Concordances

There are a number of concordances used to standardize industries across the various

trade data. The final industry classifications were driven by those available in the PNAD.

The concordances used were:

    1. ERP industry (i.e. the industry classification used by Kume et al., which is similar

        to Nivel 50) to our final industry classification (author constructed).

    2. ERP to PNAD (author constructed, based on Pavcnik et al. (2004)).

    3. Nivel 80 to Nivel 100 (available from Muendler).

    4. Nivel 100 to ERP (author constructed).

    5. SITC2 to ERP (author constructed).

The steps taken to standardize each trade variable's industry classifications are listed in

the table below.


Table A1: Industry Standardization: Steps and Concordances

Trade Variable       Initial Industry Classification   Concordance Steps   Concordances Used
Trade-weighted RER   SITC2                             SITC2 to ERP        SITC2-ERP (authors)
                                                       ERP to Final (PNAD) ERP-PNAD (authors)
ERP                  Unknown (referred to here as ERP) ERP to Final (PNAD) ERP-PNAD (authors)
Import penetration   Nivel 80                          Niv 80 to Niv 100   Niv80-Niv100 (Muendler)
                                                       Niv 100 to ERP      Niv100-ERP (authors)
                                                       ERP to Final (PNAD) ERP-PNAD (authors)
Export share         Nivel 80                          Niv 80 to Niv 100   Niv80-Niv100 (Muendler)
                                                       Niv 100 to ERP      Niv100-ERP (authors)
                                                       ERP to Final (PNAD) ERP-PNAD (authors)